inconsistent strategies to spin up models in cmip5 ... · rent generation biogeochemical models...

Geosci. Model Dev., 9, 1827–1851, 2016https://doi.org/10.5194/gmd-9-1827-2016© Author(s) 2016. This work is distributed underthe Creative Commons Attribution 3.0 License.

Inconsistent strategies to spin up models in CMIP5: implications forocean biogeochemical model performance assessmentRoland Séférian1, Marion Gehlen2, Laurent Bopp2, Laure Resplandy3,2, James C. Orr2, Olivier Marti2,John P. Dunne4, James R. Christian5, Scott C. Doney6, Tatiana Ilyina7, Keith Lindsay8, Paul R. Halloran9,Christoph Heinze10,11, Joachim Segschneider12, Jerry Tjiputra11, Olivier Aumont13, and Anastasia Romanou14,15

1CNRM, Centre National de Recherches Météorologiques, Météo-France/CNRS, 42 Avenue Gaspard Coriolis,31057 Toulouse, France2LSCE/IPSL, Laboratoire des Sciences du Climat et de l’Environnement, Orme des Merisiers,CEA/Saclay 91198 Gif-sur-Yvette CEDEX, France3Scripps Institution of Oceanography, UCSD, La Jolla, CA, USA4Geophysical Fluid Dynamics Laboratory, NOAA, Princeton, NJ, USA5Fisheries and Oceans Canada and Canadian Centre for Climate Modelling and Analysis, Victoria, B.C., Canada6Marine Chemistry and Geochemistry Department, Woods Hole Oceanographic Institution, Woods Hole, MA, USA7Max Planck Institute for Meteorology, Bundesstraße 53, 20146 Hamburg, Germany8Climate and Global Dynamics Division, National Center for Atmospheric Research, Boulder, CO, USA9College of Life and Environmental Sciences, University of Exeter, Exeter, EX4 4RJ, UK10Geophysical Institute, University of Bergen, Bergen, Norway11Uni Research Climate, Bjerknes Centre for Climate Research, Bergen, Norway12Department of Geosciences, University of Kiel, Kiel, Germany13Sorbonne Universités (UPMC, Univ Paris 06)-CNRS-IRD-MNHN, LOCEAN-IPSL Laboratory,4 Place Jussieu, 75005 Paris, France14Dept. of Applied Math. and Phys., Columbia University, 2880 Broadway, New York, NY 10025, USA15NASA-Goddard Institute for Space Studies at Columbia University, New York, NY, USA

Correspondence: Roland Séférian ([email protected])

Received: 31 August 2015 – Discussion started: 13 October 2015Revised: 19 April 2016 – Accepted: 22 April 2016 – Published: 12 May 2016

Abstract. During the fifth phase of the Coupled Model Inter-comparison Project (CMIP5) substantial efforts were madeto systematically assess the skill of Earth system models.One goal was to check how realistically representative ma-rine biogeochemical tracer distributions could be reproducedby models. In routine assessments model historical hind-casts were compared with available modern biogeochemi-cal observations. However, these assessments considered nei-ther how close modeled biogeochemical reservoirs were toequilibrium nor the sensitivity of model performance to ini-tial conditions or to the spin-up protocols. Here, we explorehow the large diversity in spin-up protocols used for marinebiogeochemistry in CMIP5 Earth system models (ESMs)contributes to model-to-model differences in the simulated

fields. We take advantage of a 500-year spin-up simulation ofIPSL-CM5A-LR to quantify the influence of the spin-up pro-tocol on model ability to reproduce relevant data fields. Am-plification of biases in selected biogeochemical fields (O2,NO3, Alk-DIC) is assessed as a function of spin-up dura-tion. We demonstrate that a relationship between spin-up du-ration and assessment metrics emerges from our model re-sults and holds when confronted with a larger ensemble ofCMIP5 models. This shows that drift has implications forperformance assessment in addition to possibly aliasing es-timates of climate change impact. Our study suggests thatdifferences in spin-up protocols could explain a substantialpart of model disparities, constituting a source of model-to-model uncertainty. This requires more attention in future

Published by Copernicus Publications on behalf of the European Geosciences Union.

1828 R. Séférian et al.: Inconsistent strategies to spin up models in CMIP5

model intercomparison exercises in order to provide quan-titatively more correct ESM results on marine biogeochem-istry and carbon cycle feedbacks.

1 Introduction

1.1 Context

Earth system models (ESMs) are recognized as the currentstate-of-the-art global coupled models used for climate re-search (e.g., Hajima et al., 2014; IPCC, 2013). They expandthe numerical representation of the climate system used dur-ing the 4th IPCC assessment report (AR4) that was limited tocoupled physical general circulation models, to the inclusionof biogeochemical and biophysical interactions between thephysical climate system and the biosphere. The ESMs thatcontributed to CMIP5 substantially differed from each otherin terms of their simulations of physical and biogeochemicalcomponents of the Earth system. These differences in designtranslate into a significant variability between the skill withwhich the different models reproduce the observed biogeo-chemistry and carbon cycle, which in turn may impact pro-jected climate change responses (IPCC, 2013).

In the typical objective evaluation and intercomparison ofthese models, a suite of standardized statistical metrics (e.g.,correlation, root-mean-squared errors) are applied to quan-tify differences between modeled and observed variables(e.g., Doney et al., 2009; Rose et al., 2009; Stow et al., 2009;Romanou et al., 2013, 2014). With the goal of constrain-ing future projections, statistical metrics are often used formodel ranking (e.g., Anav et al., 2013), weighting of modelprojections (e.g., Steinacher et al., 2010) or selection of themost skillful models across a wider ensemble (e.g., Cox etal., 2013; Massonnet et al., 2012; Wenzel et al., 2014). Mostof these approaches can be considered as “blind” given thatthey are routinely applied without considering models’ spe-cific characteristics and treat models a priori as equivalentlyindependent of observations. However, since these modelsare typically initialized from observations, the spin-up pro-cedure (e.g. the length of time for which the model has beenrun since initialization, and therefore the degree to which ithas approached its own equilibrium) has the potential to ex-ert a significant control over the statistical metrics calculatedfor each model, using those observations.

1.2 Initialization of biogeochemical fields and spin-upprotocols in CMIP5

Ocean initialization protocols aim at obtaining stable andequilibrated distributions of model state variables, such astemperature or concentrations of dissolved tracers. Mostcommonly used initialization protocols consist of initializ-ing both physical and biogeochemical variables from eitherclimatologies (derived from the observed fields or previous

model simulations) or spatially constant values before run-ning the model to equilibrium. In theory, equilibrium cor-responds to steady-state and, hence, temporal derivatives oftracer fields close to zero. The time needed to equilibratetracer distributions or, in other words, the integration timeneeded by the model to converge towards its own attractor(which is different from the true state of the climate sys-tem) varies greatly between components of the climate sys-tem. It spans from several weeks for the atmosphere (e.g.,Phillips et al., 2004) to several centuries for ocean and seaice components (e.g., Stouffer et al., 2004). The equilibra-tion of ocean biogeochemical tracers across the entire watercolumn amounts to several thousands of years (e.g., Heinzeet al., 1999; Wunsch and Heimbach, 2008) and depends onthe state of background ocean circulation as well as the tur-bulent mixing and eddy stirring parameterizations (e.g., Au-mont et al., 1998; Bryan, 1984; Gnanadesikan, 2004; Mari-nov et al., 2008). The equilibration time can be different ina coupled model configuration (i.e., ocean–atmosphere gen-eral circulation models or ESMs) compared to stand-aloneclimate components due to leaks in the energy budget (Hobbset al., 2016). In practice, these simulations, called “spin-ups”,often span only several hundreds of years, at the end of whicha quasi-equilibrium state is assumed for the interior oceantracers.

The present degree of complexity and spatial as well astemporal resolution of marine biogeochemical ESM compo-nents (as well as other physical and chemical components),however, often precludes a spin-up to reach adequate equi-libration of biogeochemical tracers. This is a consequenceof the large number of state variables present in most of thecurrent generation of biogeochemical models (e.g., for eachtracer a separate advection equation has to be solved via anumerical CPU time demanding algorithm), more complexprocess descriptions (e.g., including more plankton func-tional types than before), and spatial as well as tempo-ral resolution. This number of state variables has continu-ously increased from simple biogeochemical models (e.g.,HAMOCC3, Maier-Reimer and Hasselmann, 1987) to ma-rine biodiversity models (e.g., Follows et al., 2007). Cur-rent generation biogeochemical models embedded in CMIP5ESMs contain roughly 2 to 4 times more state variables thanphysical models (e.g., atmosphere, ocean, sea-ice), whichmakes their equilibration computationally costly and diffi-cult. The initialization of biogeochemical state variables isfurther complicated by the scarcity of biogeochemical ob-servations as compared to observations of physical vari-ables (e.g., temperature, salinity). So far, three-dimensionalobservation-based climatologies exist for macro-nutrients,oxygen, dissolved carbon and alkalinity. For other tracerssuch as dissolved iron, dissolved organic carbon and biomassof the various plankton functional types data are still sparsein space and time in-spite of considerable efforts such asthe GEOTRACES program for trace elements, or MARE-DAT for biomasses of plankton functional types. The lat-

Geosci. Model Dev., 9, 1827–1851, 2016 https://doi.org/10.5194/gmd-9-1827-2016

R. Séférian et al.: Inconsistent strategies to spin up models in CMIP5 1829

ter set of variables is initialized either with constant values(e.g., global average estimates) or with output from a previ-ous model run. An additional difficulty stems from the use ofmodern climatologies to initialize the ocean state, implicitlyassuming a long-term steady state, which does not necessar-ily represent the preindustrial state of the ocean. These clima-tologies incorporate the ongoing anthropogenic perturbationof marine biogeochemical fields, be it the uptake of anthro-pogenic CO2 or the excess of nutrients inputs and pollutants(e.g., Doney, 2010). Although methods exist to remove theanthropogenic perturbation from some observed ocean car-bon tracer fields, their use is still debated since they leadto non-unique results (e.g., Tanhua et al., 2007; Yool et al.,2010).

The equilibration of marine biogeochemical tracer dis-tributions is driven not only by the ocean circulation, butalso by numerous internal biogeochemical processes actingat various timescales. For example, while the transport anddegradation of sinking organic matter spans days to per-haps several months, the associated impact on deep waterchemistry accumulates over several decades to centuries aszones of differential remineralization are mixed across wa-ter masses and follows the ocean circulation (Wunsch andHeimbach, 2008). For models including interactive sedi-ment modules, the sediment equilibration takes even longer(O(104) years; e.g., Archer et al., 2009, and Heinze et al.,1999). As a consequence of the interplay between oceancirculation and biogeochemical processes, biogeochemicalmodels require long spin-up times to equilibrate (e.g., Khati-wala et al., 2005; Wunsch and Heimbach, 2008). Modelingstudies of paleo-oceanographic passive tracers such as δ18Oor114C (Duplessy et al., 1991), or global ocean passive trac-ers (Wunsch and Heimbach, 2008), as well as more recentlyavailable modern global-scale data compilations (e.g., Key etal., 2004; Sarmiento and Gruber, 2006) and GEOTRACESIntermediate Data product 2014 (Mawji et al., 2015) pro-vide an estimate of the time required for the ocean bio-geochemical reservoir to equilibrate with the climate sys-tems (excluding continental weathering and reaction withmarine sediments). For the deep water masses, this time isabout 1500 years in the Atlantic Ocean and reaches up to10 000 years in the North Pacific Ocean (Wunsch and Heim-bach, 2008).

In a context of model-to-model intercomparison, this timerange contributes to the model uncertainty. Lessons fromthe previous Ocean Carbon Model Intercomparison Projectphase 2 (OCMIP-2) exercise have demonstrated that somemodels required ∼ 10 000 years to reach a state where theglobal sea–air carbon flux is about 0.01 Pg C y−1.

While it is recognized that long timescale processes influ-ence the length of spin-up to equilibrium, the spin-up dura-tion is usually defined ad hoc based on external constraintsor internal biogeochemical criteria. The computational costis commonly invoked as external constraint to shorten andlimit the spin-up duration. It is directly related to model com-

plexity (e.g., Tjiputra et al., 2013; Vichi et al., 2011; Yool etal., 2013) and spatial resolution (Ito et al., 2010). The internalbiogeochemical criteria applied to derive the duration of thespin-up simulations are generally defined by (i) reaching asteady-state, quasi equilibrium of the long-term global CO2fluxes between the ocean and the atmosphere (e.g., Dunneet al., 2013; Ilyina et al., 2013; Lindsay et al., 2014; Ro-manou et al., 2013; Séférian et al., 2013), (ii) determining theamount of carbon stored in the ocean at preindustrial state(e.g., Dunne et al., 2013; Vichi et al., 2011) or (iii) repre-senting relevant biogeochemical tracer patterns (e.g., oxygenminimum zone in Ito and Deutsch, 2013).

Despite its importance, only limited information on spin-up procedures is available through the CMIP5 metadata por-tal (https://search.es-doc.org). Information on spin-up pro-tocols and model initialization is usually not taken into ac-count in model intercomparison studies (e.g., Andrews etal., 2013; Bopp et al., 2013; Cocco et al., 2013; Frölicheret al., 2014; Gehlen et al., 2014; Keller et al., 2014; Resp-landy et al., 2013, 2015; Rodgers et al., 2015; Séférian etal., 2014). This information, if available, can only be foundseparately in the reference papers of individual models (e.g.,Adachi et al., 2013; Arora et al., 2011; Collins et al., 2011;Dunne et al., 2013; Ilyina et al., 2013; Lindsay et al., 2014;Romanou et al., 2013; Séférian et al., 2013, 2016; Tjipu-tra et al., 2013; Vichi et al., 2011; Volodin et al., 2010;Watanabe et al., 2011; Wu et al., 2013). The duration ofspin-up simulations of CMIP5 ocean biogeochemical com-ponents spans from one hundred years (e.g., CMCC-CESM)to several thousand years (e.g., MPI-ESM-LR, MPI-ESM-MR) (Fig. 1 and Table 1). Model initialization and spin-upprocedures are equally variable across the model ensemble(Fig. 1 and Table 1). Four different sources of initializationand four different procedures of model equilibration emergefrom the 24 ESMs reviewed for this study.

Biogeochemical state variables were mostly initializedfrom observations, although from various releases of thesame World Ocean Atlas global climatology (WOA1994,WOA2001, WOA2006, WOA2010). A small subset of ESMsrelied either on a mix between previous model output andobservations or solely on model output from a previous sim-ulation for initialization. Similarly, spin-up procedures fallinto two categories. The first one may be called “sequen-tial”: it consists in decomposing the spin-up integration intoone long offline simulation (∼ 200–10 000 years) and oneshorter online simulation (∼ 100–1000 years). During theoffline simulation, the biogeochemical model is forced bydynamical fields from the climate model or from reanaly-sis (CanESM2, MRI-ESM, Fig. 1 and Table 1). Some mod-eling groups have adopted a “direct” strategy, which con-sists in running solely one online or coupled spin-up simula-tion (e.g., CNRM-ESM1, GFDL-ESM2M, GFDL-ESM2G,GISS-E2-H-CC, GISS-E2-R-CC, NorESM1-ME). Finally, aspin-up “acceleration” procedure is used by CMCC-CESM.This technique consists of enhancing the ocean carbon out-

https://doi.org/10.5194/gmd-9-1827-2016 Geosci. Model Dev., 9, 1827–1851, 2016

https://search.es-doc.org

1830 R. Séférian et al.: Inconsistent strategies to spin up models in CMIP5Le

ngth

of S

pin-

up (y

ear)

010

0030

0050

0070

0090

0011

000

OnlineOfflineOffline w/ accel.Forced w/ obs.

ObservationsModelMixedFixed values

BCC

−CSM

1−1

BCC

−CSM

1−1−

m

Can

ESM

2

CES

M1−

BGC

CM

CC

−CES

M

CN

RM

−CM

5

CN

RM

−CM

5−2

CN

RM

−ESM

1

GFD

L−ES

M2G

GFD

L−ES

M2M

GIS

S−E2

−H−C

C

GIS

S−E2

−R−C

C

Had

GEM

2−C

C

Had

GEM

2−ES

INM

CM

4

IPSL

−CM

5A−L

R

IPSL

−CM

5A−M

R

IPSL

−CM

5B−L

R

MIR

OC

−ESM

MIR

OC

−ESM

−CH

EM

MPI

−ESM

−LR

MPI

−ESM

−MR

MR

I−ES

M1

Nor

ESM

1−M

E

Figure 1. Spin-up protocols of CMIP5 Earth system models. Color shading represents strategies of the various modeling groups. Online andOffline steps refer to runs performed with coupled climate model and with stand-alone ocean biogeochemistry model, respectively. Sourcesof initial conditions for biogeochemical components of CMIP5 Earth system models are indicated as hatching below the barplot.

gassing to remove anthropogenic carbon from the ocean, alegacy from initialization with modern data (Global DataAnalysis Project or GLODAP following Key et al., 2004).None of these spin-up procedures, durations and sources ofinitialization can be considered as “standard”; each of themis unique and subjectively employed by one modeling group.

Objective arguments and hypotheses justifying the choiceof one method of spin-up rather than the others have been thefocus of previous studies (e.g., Dunne et al., 2013; Heinzeand Ilyina, 2015; Tjiputra et al., 2013). Similarly, individualmodeling groups have discussed the impacts of their partic-ular spin-up procedure on model performance individually(e.g., Dunne et al., 2013; Lindsay et al., 2014; Séférian et al.,2013; Vichi et al., 2011). However, no study has addressedthe potential for the large diversity of spin-up proceduresfound across the CMIP5 ensemble to translate into model-to-model differences in terms of comparative model perfor-mance assessments or model evaluations in terms of futureprojections.

1.3 Objectives of this study

This study assesses the role of the spin-up protocol in con-trolling the “final” representation of biogeochemical fields,and subsequent model skill assessment, providing a comple-mentary analysis from the studies of Sen Gupta et al. (2012,2013). It relies on a 500-year long spin-up simulation from astate-of-the-art Earth system model, IPSL-CM5A-LR to in-vestigate the impacts of spin-up strategy on selected biogeo-chemical tracers and residual model drift across the variousESMs of the CMIP5 ensemble. We demonstrate that the du-ration of the spin-up has implications for the determinationof robust and meaningful skill-score metrics that should im-prove future intercomparison studies such as CMIP6 (Meehlet al., 2014).

Section 2 describes the model, the observations, the modelexperiments, as well as the methods used for assessing theimpacts of spin-up protocols on the representation of bio-geochemical fields in IPSL-CM5A-LR, as well as across theensemble of CMIP5 ESMs. Section 3 presents the analysis



Table 1. Summary of spin-up strategy, sources of initial conditions, offline/online durations and references used to equilibrate ocean biogeo-chemistry in CMIP5 ESMs. The so-called direct and sequential strategies inform whether the spin-up of the ocean biogeochemical modelis run directly in online/coupled mode or first in offline (ocean biogeochemistry only) and then in online/coupled mode. DIC∗ refers to theobservation-derived estimates of preindustrial dissolved inorganic carbon concentration using the 1C∗ method. w/acc. and forced w/obs.indicate the strategy using “acceleration” and observed atmospheric forcings during the spin-up, respectively.

Models Spin-up Initial Offline Online Total spin-up Referencesprocedure conditions time time duration

BCC-CSM1-1 Sequential WOA2001, GLODAP 200 100 300 Wu et al. (2013)BCC-CSM1-1-m Sequential WOA2001, GLODAP 200 100 300 Wu et al. (2013)CanESM2 Sequential OCMIP profiles, CanESM1 6000 600 6600 Arora et al. (2011)

(forced w/obs.)CESM1-BGC Direct CCSM4 0 1000 1000 Lindsay et al. (2014)CMCC-CESM Sequential WOA2001, GLODAP 100 100 200 Vichi et al. (2011)

(w/acc.)CNRM-CM5 Sequential WOA1994, GLODAP, IPSL 3000 100 3100 Séférian et al. (2013)CNRM-CM5-2 Sequential WOA1994, GLODAP, CNRM 3000 100 3100 Schwinger et al. (2014)CNRM-ESM1 Sequential CNRM-CM5 0 1300 1300 Séférian et al. (2016)GFDL-ESM2G Direct WOA2005, GLODAP 0 1000 1000 Dunne et al. (2013)GFDL-ESM2M Direct WOA2005, GLODAP 0 1000 1000 Dunne et al. (2013)GISS-E2-H-CC Direct WOA2005, GLODAP DIC∗ 0 3300 3300 Romanou et al. (2013)GISS-E2-R-CC Direct WOA2005, GLODAP DIC∗ 0 3300 3300 Romanou et al. (2013)HadGEM2-CC Sequential HadCM3LC, WOA2011 400 100 500 Collins et al. (2011),

Wassmann et al. (2010)HadGEM2-ES Sequential HadCM3LC, WOA2010 400 100 500 Collins et al. (2011)INMCM4 Sequential Uniform DIC 3000 200 3200 Volodin et al. (2010)IPSL-CM5A-LR Sequential WOA1994, GLODAP, IPSL 3000 600 3600 Séférian et al. (2013)IPSL-CM5A-MR Sequential WOA1994, GLODAP, IPSL 3000 300 3300 Dufresne et al. (2013)IPSL-CM5B-LR Sequential IPSL-CM5A-LR 0 300 300 Dufresne et al. (2013)MIROC-ESM Sequential GLODAP/constant values 1245 480 1725 Watanabe et al. (2011)MIROC-ESM-CHEM Sequential GLODAP/constant values 1245 484 1729 Watanabe et al. (2011)MPI-ESM-LR Sequential HAMOCC/constant values 10 000 1900 11 900 Ilyina et al. (2013)MPI-ESM-MR Sequential HAMOCC/constant values 10 000 1500 11 500 Ilyina et al. (2013)MRI-ESM1 Sequential GLODAP 550 395 945 Adachi et al. (2013)

(forced w/obs.)NorESM Direct WOA2010, GLODAP 0 900 900 Tjiputra et al. (2013)

developed for the assessment of the impact of spin-up dura-tion on the representation of biogeochemical structures. Im-plications and recommendations are discussed in Sects. 4 and5, respectively.

2 Methods

2.1 Model simulations

This study exploits in particular results from one simula-tion performed with IPSL-CM5A-LR (Dufresne et al., 2013),considered here to be representative of the likely behavior ofother CMIP5 Earth system models. Like other current gen-eration of ESMs, IPSL-CM5A-LR combines the major com-ponents of the climate system (Chap. 9, Table 9.1 of IPCC,2013). The atmosphere is represented by the atmosphericgeneral circulation model LMDZ (Hourdin et al., 2006) witha horizontal resolution of 3.75◦× 1.87◦ and 39 levels. Theland surface is simulated with ORCHIDEE (Krinner et al.,

2005). The oceanic component is NEMOv3.2 in its ORCA2global configuration (Madec, 2008). It has a horizontal res-olution of about 2◦ with enhanced resolution at the equa-tor (0.5◦) and 31 levels. NEMOv3.2 includes the sea-icemodel LIM2 (Fichefet and Maqueda, 1997), and the marinebiogeochemistry model PISCES (Aumont and Bopp, 2006).PISCES simulates the biogeochemical cycles of oxygen, car-bon and the main nutrients with 24 state variables. Themodel simulates dissolved inorganic carbon and total alkalin-ity (carbonate alkalinity+ borate+water) and the distribu-tions of macronutrients (nitrate and ammonium, phosphate,and silicate) and the micronutrient iron. PISCES representstwo sizes of phytoplankton (i.e., nanophytoplankton and di-atoms) and two zooplankton size classes: microzooplank-ton and mesozooplankton. PISCES simulates semi-labile dis-solved organic matter, and small and large sinking particleswith different sinking speeds (3 and 50 to 200 m day−1, re-spectively). While fixed elemental stoichiometric C : N : P-1O2 ratios after Takahashi et al. (1985) are imposed for



these three compartments the internal concentrations of iron,silica and calcite are simulated prognostically. The carbonsystem is represented by dissolved inorganic carbon, alka-linity and calcite. Calcite is prognostically simulated follow-ing Maier-Reimer (1993) and Moore et al. (2002). Alkalinityin the model system includes the contribution of carbonate,bicarbonate, borate, protons, and hydroxide ions. Oxygen isprognostically simulated. The model distinguishes betweenoxic and suboxic remineralization pathways, the former re-lying on oxygen as electron acceptor, the latter on nitrate.For carbon and oxygen pools, air-sea exchange follows theWanninkhof (1992) formulation.

The model’s boundary conditions account for nutrient sup-plies from three different sources: atmospheric dust depo-sition for iron, phosphorus and silica (Jickells and Spokes,2001; Moore et al., 2004; Tegen and Fung, 1995), rivers fornutrients, alkalinity and carbon (Ludwig et al., 1996) andsediment mobilization for sedimentary iron (de Baar and deJong, 2001; Johnson et al., 1999). To ensure conservation ofnitrogen in the ocean, annual total nitrogen fixation is ad-justed to balance losses from denitrification. For the othermacronutrients, alkalinity and organic carbon, the conserva-tion is ensured by tuning the sedimental burial to the totalexternal input from rivers and dust. In PISCES, an adequatetreatment of external boundary conditions has been demon-strated to be essential for the accurate simulation of nutrientdistributions (Aumont and Bopp, 2006; Aumont et al., 2003).Riverine carbon inputs induce a natural outgassing of carbonof 0.6 Pg C y−1 that has been shown to be essential to modelthe inter-hemispheric gradient of atmospheric CO2 under apreindustrial state (Aumont et al., 2001).

The core simulation used within this study is a 500-yearlong coupled preindustrial run. It uses the same atmospheric,land surface and ocean configurations as IPSL-CM5A-LR(Dufresne et al., 2013) for which the marine biogeochem-istry has been extensively evaluated (see, e.g., Séférian et al.,2013, for modern-state evaluation). The only difference be-tween the “standard” preindustrial simulation contributed toCMIP5 and the present one is the initial conditions. Whilethe CMIP5 preindustrial simulation starts from an ocean cir-culation after several thousand years of online physical ad-justment, the present simulation starts from an ocean at restusing the January temperature and salinity fields from theWorld Ocean Atlas (Levitus and Boyer, 1994). Biogeochem-ical state variables were initialized from data compilations orclimatologies as explained in the following section. Atmo-spheric CO2 and other greenhouse gases, as well as naturalaerosols, were set to their 1850 preindustrial values. The sim-ulation is extensively described in terms of ocean physics byMignot et al. (2013). Mignot and coworkers show that thestrength of the Atlantic meridional overturning circulationand the Antarctic circumpolar current as well as the upper300 m ocean heat content stabilize after 250 years of simula-tion.

Although the spin-up protocol used to conduct this 500-year long simulation is not readily comparable to the oneused to produce the initial conditions for the CMIP5 prein-dustrial simulation, its duration is greater than the medianlength of online adjustment computed from the multiplespin-up protocols applied during CMIP5 (∼ 395 years, Fig. 1and Table 1). Besides, the methodology of initializing bio-geochemical state variables from data fields is not broadlyemployed by the various modeling groups that have con-tributed to CMIP5. Despite the above-mentioned method-ological shortcuts, we take this 500-year long preindustrialsimulation as a representative example of a spin-up protocolgiven the diversity of approaches used by CMIP5 models.

2.2 Observations for initialization and evaluation

Two streams of data sets were used in this study. The firststream combines data from the World Ocean Atlas 1994(WOA94, Levitus and Boyer, 1994, and Levitus et al., 1993)for the initialization of three-dimensional fields of temper-ature and salinity, dissolved nitrate, silicate, phosphate andoxygen, and data from GLODAP (Key et al., 2004) for prein-dustrial dissolved inorganic carbon and total alkalinity. Thisstream of data was chosen purposely in our experimental set-up to be slightly different than the second stream of data,World Ocean Atlas 2013 (WOA2013, Levitus et al., 2013),the evaluation data set.

A second stream of data was used to compare modeledbiogeochemical fields. It includes up-to-date observed cli-matologies of nitrate and oxygen from the WOA2013. Thisdatabase is based on samples collected since 1965, and in-cluding data more recently collected than that made useof in WOA94. For the concentrations of preindustrial dis-solved inorganic carbon and total alkalinity, we still useGLODAP. The second stream of data was selected to be asclose as possible to the “standard” evaluation procedure ofskill-assessment protocols found in CMIP5 model referencepapers (Adachi et al., 2013; Arora et al., 2011; Collins etal., 2011; Dunne et al., 2013; Ilyina et al., 2013; Lindsay etal., 2014; Romanou et al., 2013; Séférian et al., 2013, 2016;Tjiputra et al., 2013; Vichi et al., 2011; Volodin et al., 2010;Watanabe et al., 2011; Wu et al., 2013). Differences betweenthese two streams of data are minor and are further detailedbelow.

2.3 Approach and statistical analysis

To quantify the impacts of a large diversity of spin-up pro-cedures on the representation of biogeochemical fields inCMIP5, we employ a three-fold approach.

1. The 500-year long spin-up simulation described inSect. 2.1 is used to determine the influence of the spin-up procedure on the representation of biogeochemicalfields in IPSL-CM5A-LR.



2. In the next step, relationships between biases in mod-eled fields, model–data mismatches and the duration ofthe spin-up simulation are identified across the CMIP5ensemble. For this step, drifts in biogeochemical fieldsare determined from the first century of the preindustrialsimulation (referred to as piControl) of each CMIP5ESM.

3. Finally, the ensembles of industrial-revolution topresent-day simulation (referred to as historical) fromeach available CMIP5 ESM are used to estimate theimpact of these drifts in biogeochemical fields on theability of models to replicate modern observations. Fora given model, we use the ensemble average of theavailable “historical” members if several realizationsare available.

For this purpose, several statistical skill score metrics arecomputed following Rose et al. (2009) and Stow et al. (2009)from model fields interpolated on a regular 1◦ grid and tofixed depth levels. The skill score metrics are (1) the glob-ally averaged concentrations for overall drift; (2) the error orbias between modeled and observed fields at each grid cell;(3) spatial correlation between model and observations to as-sess mismatches between modeled and observed large-scalestructures; (4) the root-mean squared error (RMSE) to assessthe total cumulative errors between modeled and observedfields. These statistical metrics are computed at differentdepth levels, but for clarity we focus on surface, 150 m (ther-mocline) and 2000 m (deep) levels. These statistical metricswere chosen among those described in the literature, becausethey proved to yield the most indicative scores for trackingmodel errors or improvement along the various intercompar-ison exercises (IPCC, 2013).

The drift is determined for either concentrations in simu-lated biogeochemical fields or for skill score metrics (e.g.,RMSE) using a linear regression fit over a time windowof 100 years. This time window of 100 years was chosenas a trade-off between a longer time window (> 200 years)that smoothes the drift signal and a shorter time window(< 100 years) that introduces fluctuations due to internal vari-ability and hence impacting the quality of the fit (see the as-sessment performed with the millennial-long CMIP5 piCon-trol simulation of IPSL-CM5A-LR in Fig. S1 in the Supple-ment).

The drift is assumed to decrease exponentially during thespin-up simulation and is described by a simple drift model:

drift(t)= drift(t = 0)× exp(−1τt) (1)

where τ is the relaxation time of the respective field at a givendepth level. It corresponds to the time required to nullify thedrift.

Our analyses focus on the global distribution of nitrate(NO3), dissolved oxygen (O2) and the difference between to-tal alkalinity and dissolved inorganic carbon (Alk-DIC).

The latter serves as an approximation of carbonate ionconcentration following Zeebe and Wolf-Gladrow (2001).We use this approximation of the carbonate ion concentra-tion rather than its concentration, [CO2−

3 ], since the latterwas poorly assessed in CMIP5 reference papers and wasnot provided by a majority of ESMs. These three biogeo-chemical tracers were chosen because (1) most current bio-geochemical models simulate Alk, DIC, NO3 and O2 prog-nostically and (2) they are frequently used in state-of-the-art model performance assessment (e.g., Anav et al., 2013;Bopp et al., 2013; Doney et al., 2009; Friedrichs et al., 2009,2007; Stow et al., 2009), and (3) DIC and Alk are both usedas “master tracers” for the carbonate system in the oceanbiogeochemistry models (while [CO2−

3 ], e.g., is not explic-itly transported as a tracer with the velocity fields but diag-nosed from temperature, salinity, DIC, Alk, [H+], and pCO2when needed). Modeled distributions of NO3, O2 and Alk-DIC reflect the representation of biogeochemical processesrelated to the biological pump (CO2, NO3, O2), the air-seagas exchange and ocean ventilation (CO2 and O2), as well ascarbonate chemistry (Alk-DIC). These biogeochemical pro-cesses are of particular relevance for investigating the impactof climate change on marine productivity (e.g., Henson et al.,2010), ocean deoxygenation (e.g., Gruber, 2011; Keeling etal., 2009) and the ocean carbon sink, processes for which fu-ture projections with the current generation of ESMs yieldlarge inter-model spreads (e.g., Friedlingstein et al., 2013;Resplandy et al., 2015; Séférian et al., 2014; Tjiputra et al.,2014).

3 Results

3.1 Comparison of observational data sets

Our review of spin-up protocols for CMIP5 ESM shows thatseveral modeling groups have employed different streamsof data sets to initialize their biogeochemical models (e.g.,WOA1994, WOA2001), while model evaluation relies onthe most up-to-date stream of data. To investigate the dif-ferences between the two data streams used for initializingand assessing, respectively, NO3 and O2 concentrations areanalyzed. Table 2 summarizes the RMSE and correlation be-tween WOA1994 and WOA2013 for these two biogeochem-ical fields.

Table 2 indicates that differences between the two streamsof data are fairly small. The total difference (RMSE) repre-sents a departure between 5 to 10 % from the global averageconcentrations of WOA2013 across depth levels. It is gen-erally lower in regions where the sampling density has notincreased markedly between the two releases. These valuescan be used as a baseline for model-to-model comparisonassuming that errors attributed to the various sources of ini-tialization cannot be larger than 10 %. Considering that somemodels have used outputs from previous model simulations



Table 2. Differences between the oxygen (O2, µmol L−1) and nitrate (NO3, µmol L−1) data sets used for initializing IPSL-CM5A-LR(WOA1994) and the data sets used for assessing its performance (WOA2013).

O2 NO3

Depth Surface 150 m 2000 m Surface 150 m 2000 m

RMSE 7.19 8.75 5.50 2.07 2.90 2.08R2 0.98 0.98 0.99 0.96 0.92 0.94

or globally averaged concentrations as initial conditions, weacknowledge that this baseline is not a perfect criterion forbenchmarking model performance. There is, however, noideal solution to address this issue since there is no stan-dardized set of initial conditions in CMIP5 except some rec-ommendations for the decadal prediction exercise in whichspecific attention was paid to initialization (e.g., Keenlysideet al., 2008; Kim et al., 2012; Matei et al., 2012; Meehl etal., 2013, 2009; Servonnat et al., 2014; Smith et al., 2007;Swingedouw et al., 2013).

3.2 Equilibration state metrics in IPSL-CM5A-LR

The global mean sea surface temperature (SST) is a com-mon metric to quantify the energetic equilibrium of themodel. This metric has been widely used in various papersreferenced in this study to determine the equilibration ofESM physical components. Figure 2a shows the evolutionof this metric during the 500-year long spin-up simulation.The global average SST sharply decreases during the first250 years of the simulation. In the last 250 years of the simu-lation, the global averaged SST displays a small residual driftof ∼−10−4 ◦C y−1, which falls into the range of the driftsreported for CMIP5 ESMs (Sen Gupta et al., 2013). The evo-lution over the last 250 years is comparable to those of otherphysical equilibration metrics, such as the ocean heat con-tent or the meridional overturning circulation (Mignot et al.,2013).

To assess if ocean carbon cycle reservoirs are equili-brated, we track the temporal evolution of sea-to-air CO2fluxes during the spin-up simulation. This metric was used inphase 2 of the Ocean Carbon Model Intercomparison Project(OCMIP-2; Orr, 2002) and has still widely been used dur-ing CMIP5 as an equilibration metric for the marine biogeo-chemistry. Figure 2b presents its evolution in the 500-yearlong spin-up simulation. The global ocean sea-to-air CO2flux is ∼−0.7 Pg C y−1 over the last decades of the spin-upsimulation (negative values indicate ocean CO2 uptake).

We use the range of values estimated from preindustrialnatural ocean carbon flux inversions (e.g. Gerber and Joos,2010, or Mikaloff Fletcher et al., 2007) to evaluate the globalsea-to-air carbon flux simulated by IPSL-CM5A-LR. Since,these estimates do not account for the preindustrial car-bon outgassing induced by the river input, while our modeldoes, we have added a constant outgassing of 0.45 Pg C y−1

to the range of 0.03± 0.08 Pg C y−1 (Mikaloff Fletcher etal., 2007). This value of 0.45 Pg C y−1 corresponds to theglobal open-ocean river-induced carbon outgassing accord-ing to IPCC (2013) or Le Quéré et al. (2015). Consequently,in our modeling framework, the target value of the globalsea-to-air carbon flux ranges between 0.4 and 0.56 Pg C y−1.

Figure 2b shows that the global sea-to-air carbon fluxis still lower than the range of values estimated frompreindustrial natural ocean carbon flux inversions (∼ 0.4–0.56 Pg C y−1). Besides, Fig. 2b shows that the drift in theglobal sea-to-air carbon flux becomes smaller more slowlyafter a strong decline during the first 50 years of thespin-up simulation. From year 250–500 this drift is about0.001 Pg C y−2 and still weaker over the last century of thesimulation (7× 10−4 Pg C y−2). A one-sided t-test indicatesthat the two drifts differ from each other with a p value < 2×10−16. When fitted with drifts computed from overlappingtime segments of 100 years, our simple drift model (Eq. 1)gives a relaxation time of around 160 years. We use this re-laxation time and the drift of 7× 10−4 Pg C y−2 to estimatethe additional spin-up time required for the model to reachan outgassing of 0.4–0.56 Pg C y−1, as 1100 to 1300 years.However, even after this integration time, the drift in globalsea-to-air carbon flux estimated with our simple drift modelstill ranges from 2× 10−7 to 7× 10−7 Pg C y−2.

These estimates do not account for the non-linearity of theocean carbon cycle and the associated process uncertainties(Schwinger et al., 2014), and hence potentially underestimatethe time required to equilibrate the ocean carbon cycle andsea-to-air carbon fluxes in the range of inversion estimates.The drift of 0.001 Pg C y−2 is, however, much smaller thanthe oceanic sink for anthropogenic carbon. Even if not fullyequilibrated in terms of carbon balance, it is likely that thisrun would have given consistent estimates of anthropogeniccarbon uptake in transient historical hindcasts.

3.3 Temporal evolution of model errors inIPSL-CM5A-LR

Figure 3 shows the temporal evolution of globally aver-aged concentrations for O2, NO3 and Alk-DIC at the sur-face (Fig. 3a, b and c), 150 m (Fig. 3d, e and f) and 2000 m(Fig. 3g, h, and i). Globally averaged concentrations of O2,NO3 and Alk-DIC (solid lines) reach steady state after 100 to250 years of spin-up at the surface. While modeled nominal



Sea

surfa

ce te

mpe

ratu

re [°

C]

(a)

17

.01

7.5

18

.0

Time [yr]

Oce

an c

arbo

n flu

x [P

g C

y-1]

(b)

1 50 100 150 200 250 300 350 400 450 500

−3

−2

−1

01

Figure 2. Time series of two climate indices over the 500-year spin-up simulation of IPSL-CM5A-LR. They represent the globally averagedsea surface temperature (a) and the global mean sea-air carbon flux (b). For sea-air carbon flux, negative value indicates uptake of carbon.Steady state equilibrium of physical components as described in Mignot et al. (2013) is reached at∼ 250 years and is indicated with a verticaldashed line. Drifts in sea surface temperature and global carbon flux are indicated with dashed blue lines. They are computed using a linearregression fit over years 250 to 500. Hatching on panel (b) represents the range of inverse modeling estimates for preindustrial global carbonflux as described in Mikaloff Fletcher et al. (2007), i.e., 0.03± 0.08 plus 0.45 Pg C y−1 corresponding to the riverine-induced natural CO2outgassing outside of near-shore regions consistently with Le Quéré et al. (2015).

values for O2 concentration converge toward the observedconcentration (i.e., 172.3 µmol L−1), that of NO3 presentspersistent deviations from WOA2013. At the surface, theconvergence of the simulated oxygen to observed values isexpected since the dominant governing process of thermody-namic saturation (through the air-sea gas exchange) is wellunderstood and modeled. The deviation in surface NO3 high-lights uncertainty related to near surface biological processesand upper ocean physics. Below the surface, concentrationsof biogeochemical tracers drift away from the globally av-eraged concentrations computed from WOA2013 or GLO-DAP (Fig. 3d–i). At 150 and 2000 m, the drift in global aver-aged concentrations for these fields, computed over the last250 years, is still significant with p < 10−4 (Table 3). Ex-cept for the surface fields, Fig. 3 shows that RMSE, indicatedwith dashed lines in Fig. 3, globally increases with time forall biogeochemical fields. The linear drift in RMSE over thelast 250 years of the spin-up simulation falls within the 2–3 % ky−1 range at the surface. It is much larger at 2000 m(144–280 % ky−1 ; Table 3). This is also the case regionally,because the latitudinal maximum in RMSE (RMSEmax) issimilar to the global RMSE. Table 3 also shows that the mag-nitude of drift in RMSE for O2, NO3 and Alk-DIC differs at

a given depth as different processes affect the interior distri-bution of these biogeochemical fields.

3.4 Evolution of geographical mismatches inIPSL-CM5A-LR

To further explore the evolution of mismatch in biogeochem-ical distributions, we analyze differences (ε) between sim-ulated and observed fields of O2 and NO3 from WOA2013and Alk-DIC from GLODAP after the initialization and atthe end of the spin-up, i.e., the first year and the last year ofthe core spin-up simulation performed with the IPSL-CM5A-LR model (Figs. 4, 5 and 6).

Figure 4a, c, e shows that surface concentrations of bio-geochemical fields are associated with small biases at initial-ization. This error represents less than 5 % of the observedsurface concentrations for O2, NO3 and Alk-DIC and reflectsthe weak difference between the data stream employed forinitialization and validation. After 500 years of spin-up, de-viations between the modeled and observed fields at the sur-face have increased locally by up to ∼ 40 % (Fig. 4b, d, andf). The largest deviations are found in high-latitude oceansfor O2 and NO3 and also to some extent in the tropics forNO3 and Alk-DIC.



O2 (μ

mo

l.L−1)

16

81

69

17

11

72

17

4

78

10

11

13

(a)N

O3 (μ

mo

l.L−1)

1.6

2.2

2.8

3.4

0.6

1.2

1.8

2.4

(b)

Alk

−D

IC (μ

mo

l.L−1)

19

42

01

.42

08

.82

16

.22

23

.62

31

10

17

.42

4.8

32

.23

9.6

47

(c)

O2 (μ

mo

l.L−1)

11

51

21

12

71

32

13

81

44

612

18

23

25

(d)

NO

3 (μ

mo

l.L−1)

55

.86

.67

.48

.29

00

.81

.62

.43

.24

(e)

Alk

−D

IC (μ

mo

l.L−1)

12

91

36

.61

44

.21

51

.81

59

.41

67

61

3.6

21

.22

8.8

36

.44

4

(f)

Time [year]

O2 (μ

mo

l.L−1)

1 100 200 300 400 500

78

83

88

92

97

49

14

18

23

(g)

Time [year]

NO

3 (μ

mo

l.L−1)

1 100 200 300 400 500

19

19

.21

9.4

19

.61

9.8

20

11

.21

.41

.61

.82

(h)

Time [year]

Alk

−D

IC (μ

mo

l.L−1)

1 100 200 300 400 500

34

39

.44

4.8

50

.25

5.6

61

16

.41

1.8

17

.22

2.6

28

(i)

Figure 3. Time series of globally averaged concentration (in solid lines) and globally averaged root-mean squared error (RMSE in dashedlines) for dissolved oxygen (O2), nitrate (NO3) and difference between alkalinity and dissolved inorganic carbon (Alk-DIC) as simulatedby IPSL-CM5A-LR. Globally averaged concentration and RMSE are given at surface (a, b, c), 150 m (d, e, f), and 2000 m (g, h, i) forthese three biogeochemical fields. Their values are indicated on the left-side and right-side y axis, respectively. Hatching represents the ±σobservational uncertainty due to optimal interpolation of in situ concentrations around the observed globally averaged concentration.



Table 3. Drift in % ky−1 for oxygen (O2), nitrate (NO3) and total alkalinity minus DIC (Alk-DIC) at surface, 150 and 2000 m as simulatedby the IPSL-CM5A-LR model. The drift has been computed over the last 250 years of the spin-up simulation using a linear regression fit ofthe globally averaged concentrations, root-mean squared error (RMSE) and latitudinal maximum root-mean squared error (RMSEmax) withrespect to the values at year 250.

O2 NO3 Alk-DIC

Metrics Mean RMSE RMSEmax Mean RMSE RMSEmax Mean RMSE RMSEmax

Surf −0.2 2.6 55.8 −0.1 −0.1 34.2 1.6 −0.1 −0.1150 m 3.4 39.0 31.5 −15.9 33.4 55.2 6.1 27.9 24.72000 m −30.4 144.3 −40.1 2 51.8 −34.8 −69.6 281.8 47.5

Figure 4. Snapshots of spatial biases, ε, in surface concentrations (µmol L−1) in biogeochemical fields during the 500-year spin-up simulationof IPSL-CM5A-LR. ε in dissolved oxygen (O2), nitrate (NO3) and difference between alkalinity and dissolved inorganic carbon (Alk-DIC)is given for the first year (a, c, e, respectively) and for the last year of spin-up simulation (b, d, f, respectively).

Below the surface, distributions of modeled biogeochem-ical fields compare well to the observations at 150 m at ini-tialization with averaged errors close to zero (Fig. 5a, c, ande). This result was expected since WOA2013 and WOA1994differ little at these depth levels. Subsurface distributions atinitialization strongly contrast with the concentrations thatresulted from 500 years of spin-up (Fig. 5b, d, and f). Af-ter 500 years of spin-up, substantial mismatches character-ize the distribution of O2, NO3 and Alk-DIC fields in thehigh-latitude oceans and in the tropics. Figure 5 illustratesthat patterns of errors for O2, NO3 and Alk-DIC fields arewell correlated with each other (R > 0.6). This reflects thatin PISCES carbon, nitrogen and oxygen concentrations arelinked by the elemental C : N : -1O2 stoichiometry fixed inspace and time. Figure 6 shows that model–data deviations at2000 m have substantially increased at a regional level after500 years of simulation, showing large errors in the SouthernHemisphere oceans. This appears clearly in Fig. 6d and f forNO3 and Alk-DIC fields, respectively.

The temporal evolution of the RMSE between modeledand observed fields of O2, NO3 and Alk-DIC over the wholewater column is presented in Fig. 7 in terms of RMSE(Fig. 7a–c). As expected, Fig. 7 illustrates that there is a goodmatch during the first years of simulation for all biogeochem-ical fields at all depth levels with low RMSE. After a fewcenturies, patterns of error evolve differently across depth forO2, NO3 and Alk-DIC.

The temporal evolution of RMSE shows that patterns of er-ror have reached a steady state a few decades after 250 yearsof spin-up within the upper hundred meters of the ocean butcontinue to evolve at greater depths, even after 500 years.Patterns of errors within the thermocline and upper 1000 mwater masses evolve relatively fast (within a few centuries)due to the relatively short mixing time in the upper ocean.Mid-depth (∼ 1500–2500 m) RMSE evolves much slowerbecause of the slow ocean circulation at these depth levels.Characteristic timescales here are thousands of years as ev-idenced by the observed radiocarbon age of seawater (e.g.,



Figure 5. As Fig. 4 but for concentrations at 150 m. Note that color shading does not represent the same amplitude in spatial biases as inFigs. 4 and 6.

Figure 6. As Fig. 4 but for concentrations at 2000 m. Note that color shading does not represent the same amplitude in spatial biases as inFigs. 4 and 5.

Wunsch and Heimbach, 2007, 2008). This explains why, atthe end of the spin-up simulation, two maxima of compara-ble amplitude are found for RMSE at 150 and 3750 m for O2and at 50 and 3800 m for Alk-DIC (Fig. 7).

3.5 Drifts in IPSL-CM5A-LR spin-up simulation

With the evolution of the RMSE established, we can usethe simple drift model (Eq. 1) to determine the relaxationtime, τ , which characterizes the e-folding timescale of the

RMSE. To use this simple drift model, we compute the driftin RMSE determined from time segments of 100 years dis-tributed evenly every 5 years from year 250 to 500 for O2,NO3 and Alk-DIC tracers. The drift model (magenta lines inFig. 8) is fitted to the 80 drift values for each field and eachdepth level (colored crosses in Fig. 8).

The simple drift model fits well the evolution of the driftin RMSE for the biogeochemical variables along the spin-upsimulation of IPSL-CM5A-LR (Fig. 8). Correlation coeffi-cients are mostly significant at 90 % confidence level (r∗ =



Figure 7. Temporal–vertical evolution in root-mean squared error (RMSE) for biogeochemical tracers during the 500-year long spin-upsimulation of IPSL-CM5A-LR. The RMSE is given for (a) dissolved oxygen O2, (b) nitrate NO3 and the (c) difference between alkalinityand dissolved inorganic carbon Alk-DIC.

0.3 determined with a student distribution with significancelevel of 90 % and ∼ 15 effective degrees of freedom esti-mated with the formulation of Bretherton et al., 1999), exceptfor NO3 at surface and Alk-DIC at 150 and 2000 m. Anotherexception is found for NO3 at 150 m where the drift does notcorrespond to an exponential decay of the drift as functionof time. The large confidence interval of the fit indicates thatthe fit would have been considered as non-significant given alonger spin-up simulation or a higher confidence threshold.

When significant, estimates of τ for O2 RMSE are ≈ 90,564 and 1149 y at the surface, 150 and 2000 m, respectively.These values match reasonably well τ estimated for NO3RMSE at 2000 m (1130 y) and those for Alk-DIC RMSE atsurface and 2000 m (137 and 1163 y). However, these esti-mates are sensitive to the time windows used to computethe drift. For a subset of time windows between 100 and250 years by step of 50 years, τ estimates for O2 RMSE are≈ 114± 67, 375± 140 and 1116± 527 y at the surface, 150and 2000 m depth. These large uncertainties associated withτ estimates are essentially due to the length of the spin-upsimulation. A longer spin-up simulation would improve thequality of the fit (see Fig. S1).

3.6 Drifts in CMIP5 ESM preindustrial simulations

In this subsection, the analysis is extended to the CMIP5archive. We focus on oxygen fields in the long preindustrialsimulation, piControl, for the 15 available CMIP5 ESMs.From these simulations that span from 250 to 1000 years,we compute the drift in O2 RMSE across depth from severaltime segments of 100 years distributed evenly every 5 yearsfrom the beginning until the end of the piControl simula-tion. These drifts are used as a surrogate for drift computedfrom the spin-up of each model since such simulations arenot available through the data portal.

Figure 9 represents the drift in O2 RMSE vs. the spin-upduration for each CMIP5 ESM. The analysis shows that the

drift in O2 RMSE differs substantially between models. For agiven model, drifts in other biogeochemical tracers (NO3 andAlk-DIC) display similar features (not shown). The between-model differences in drift are not surprising since there areno reasons for different models to exhibit similar drift fora given field. Yet, Fig. 9 shows that a global relationshipemerges from this ensemble when using the simple driftmodel to fit the drift in O2 RMSE as a function of the spin-upduration (solid green lines in Fig. 9). With a 90 % confidencelevel, this relationship suggests a general decrease in the driftas a function of spin-up duration for all depth levels. At thesurface and at 2000 m depth, the quality of fits is low, withcorrelation coefficients of about 0.4. These are however sig-nificant at the 90 % confidence level (r∗ = 0.34 determinedwith a student distribution with a significance level of 90 %and 15 models as a degree of freedom). The weakest correla-tion coefficient is found for the fit at 150 m depth and henceindicates that there is no link between the drift in O2 RMSEand the duration of the spin-up simulation. This low signif-icance level must be put into perspective given the large di-versity of spin-up protocols and initial conditions (Fig. 1 andTable 1) that can deteriorate the drift–spin-up duration rela-tionship in this ensemble of models.

The drift vs. spin-up duration relationship establishedfrom the 15 CMIP5 ESMs is nonetheless consistent with theresults obtained with IPSL-CM5A-LR (the results in Fig. 8have been reported in Fig. 9 with magenta crosses). Indeed,the drifts in RMSE decrease in the course of time at the var-ious depth levels for the IPSL-CM5A-LR model, althoughtheir magnitudes differ. This difference in magnitude is notsurprising if one considers that drift is highly model and pro-tocol dependent and that the length of the IPSL-CM5A-LRspin-up simulation is potentially too short to determine ac-curate estimates of the long-term drift in O2 RMSE. Despitethese differences, our analyses show that a relationship be-tween the drift in O2 RMSE vs. the spin-up duration emerges



+

+

+

++++++++++++++

+++++++++

+++++++++

++++++++++++++++++

+++++++++++++

++++++

++++++++0

510

1520

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)

(a)

+

+++++++++++++++

+++++++++++++++

++++++++

++++++++++++++

+++++++++

+++++++

+++++++++++

010

2030

4050

(b)

+

+

++++++++++++++++++++

+++++++++++++++++++++

++++++++++++++++++++++++++++

+++++++++

100 200 300

020

4060

80

(c)

+

+

+

++++

++++++

+++

+

++++++++

+

+

+++++

+

++++++

+

+++

+++

+

++++

+

++

++

+

+

+

+++++

++

+

+++

++

+++++

++

+

0.0

0.5

1.0

1.5

2.0

Drif

t in

NO

RM

SE (1

03

−3(d)

+

++

+

+++++++++

+

++

+

+++++++++

++++++++

+

++++++++++++++++++

++++

+

++++++++++++++++++++++

0.0

0.5

1.0

1.5

2.0

2.5

3.0

(e)

+

+++++++++++++++++++++++++++++++++

++++++++++++++++++++++++++++++++++++++++++++++

100 200 300

01

23

45

Time from year 250 of the spin-up (yr)

(f)

+

+

+

+

++++++++++++++++++++

+++++++

+++++++++++++

+++++++++++++++++++++

+++++++++++++++

010

2030

40D

rift i

n Al

k−D

IC R

MSE

(10−

3

(g)

+

+

+

+++++++++++

+++++++

+++++++++++

++++++++++

++++++

+++++++++++++++++++++

+++++++++++

010

2030

40

(h)

+++++++++++++++++

++++++++++++++++++++++++++++++++++++++++++++

+++++++++++++++++++

100 200 300

010

2030

40

(i)

R = 0.66 *

R = 0.48 *

R = 0.6 *

R = 0.02

R = 0.18

R = 0.92 *

R = 0.27 *

R = 0.03

R = 0.21

−1−1

μm

ol L

10

0yr

)−1−1

μm

ol L

10

0yr

)−1−1

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)−1−1

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)−1−1

Drif

t in

NO

RM

SE (1

03 μ

mol

L

100y

r )−1

−1D

rift i

n N

O R

MSE

(10

3 μm

ol L

10

0yr

)−1−1

Drif

t in

Alk−

DIC

RM

SE 1

0−3 μ

mol

L

100y

r )−1

−1D

rift i

n Al

k−D

IC R

MSE

10−

3 μm

ol L

10

0yr

)−1−1

Figure 8. Temporal evolution of drift in root-mean squared error (RMSE) for dissolved oxygen (O2, blue crosses), nitrate (NO3, greencrosses) and difference between alkalinity and dissolved inorganic carbon (Alk-DIC, orange crosses) during the 500-year long spin-upsimulation of IPSL-CM5A-LR. Drift in RMSE is given at surface (a, b, c), 150 m (d, e, f), and 2000 m (g, h, i) for these three biogeochemicalfields. Drift in RMSE is computed from time segments of 100 years beginning every 5 years from the beginning until year 400 of the spin-upsimulation for O2, NO3 and Alk-DIC tracers. The best-fit regressions between drifts in RMSE and spin-up duration over year 250 to 500are indicated in solid magenta lines; their 90 % confidence intervals are given by thin dashed envelopes. Least square correlation coefficientsare tested against a one-tailed t-test with significance level of 90 % and ∼ 15 effective degrees of freedom estimated with the formulation ofBretherton et al. (1999); ∗ indicates if a given least square correlation coefficient passes the test.



02

46

81

01

2

+

+

+

++

+

++

+

++

++

++

+

+++++++

+

+

+

++++

+++

++

+

+

++

++

+++

+

+

+

+

++

+

+

+

++

++

+

++

+

+

+

+

+

+++

+

+

+

++++

+

++

R = 0.44

(a)

05

10

15

20

25

30

+

+

+

+

+

+

+++

++++

+++++++++

++++++

+++

++++

+++++++

+

+

+++

++

++++

+

+++

++++

+

++++++++++++++

R = 0.14

(b)

0 2000 4000 6000 8000 10000 12000

02

04

060

80

Spin-up Duration (yr)

+

+

+

+

+

+

+

+

+

+

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

R = 0.43

(c)

CESM1−BGCCMCC−CESMCNRM−CM5CNRM−CM5−2CNRM−ESM1GFDL−ESM2GGFDL−ESM2MHadGEM2−CCHadGEM2−ESIPSL−CM5A−LRIPSL−CM5A−MRIPSL−CM5B−LRMPI−ESM−LRMPI−ESM−MRNorESM1−ME

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)−1−1

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)−1−1

Drif

t in

O2 R

MSE

(10−

3 μm

ol L

10

0yr

)−1−1

Figure 9. Scatterplot of drifts in root-mean squared error (RMSE) in O2 concentration vs. the duration of the spin-up simulation for theavailable CMIP5 Earth system models. Drifts in O2 RMSE are respectively given for surface (a), 150 m (b) and 2000 m (c) for oxygenconcentrations. Drift in O2 RMSE is computed from several time segments of 100 years beginning every 5 years from the beginning untilthe end of the piControl simulation for the available CMIP5 models. Colored symbols indicate the mean drift in O2 RMSE, while verticallines represent the associated 90 % confidence interval. The best-fit regressions between models’ mean drifts in RMSE and spin-up durationare indicated as solid green lines; their 90 % confidence intervals are given by thin dashed envelopes. Fits are assumed robust if correlationcoefficients are significant at 90 % (i.e., r∗ > 0.34). For comparison, drift in O2 RMSE from our spin-up simulation with IPSL-CM5A-LR(Fig. 8) are represented by magenta crosses.



from an ensemble of models and is broadly consistent withour theoretical framework of a drift model established fromthe results of the IPSL-CM5A-LR model (Fig. 8).

3.7 Impact of the drift on model skill score assessmentmetrics across CMIP5 ESMs

In the following, we investigate the influence of model drifton skill score assessment metrics that are routinely used tobenchmark model performance. For this purpose, we use theensemble-mean O2 RMSE as a metric to assess the distancebetween the biogeochemical observations and model results.For this purpose, we compute O2 RMSE from each ensemblemember of the CMIP5 models averaged from 1986 to 2005with respect to WOA2013 observations. The model–data dis-tance is then determined for each CMIP5 model using themean across the available ensemble members.

The left-hand side panels of Fig. 10 present the perfor-mance of available CMIP5 models in terms of distance tooxygen observations at the surface, 150 and 2000 m, respec-tively. In these panels, the various CMIP5 models are or-dered as function of their distance to the oxygen observa-tions. Following Knutti et al. (2013), either the ensemblemean or the ensemble median is used to identify groups ofmodels with similar skill within the CMIP5 ensemble. Theleft-hand side panels of Fig. 10 show that the ability of mod-els to reproduce oxygen observations varies across depth lev-els. The RMSE in the simulated O2 fields in CESM1-BGC,HadGEM2-ES, HadGEM2-CC, GFDL-ESM2M, MPI-ESM-LR and MPI-ESM-MR is generally smaller than the en-semble mean or ensemble median RMSE across the var-ious depth levels (Fig. 10a, b and c). On the other sideof the ranking, CMCC-CESM, CNRM-CM5, CNRM-CM5-2, IPSL-CM5B-LR and NorESM1-ME exhibit RMSE gen-erally higher than the ensemble mean and median RMSEacross the various depth levels. The other models, i.e.,CNRM-ESM1, GFDL-ESM2G, IPSL-CM5A-LR and IPSL-CM5A-MR display O2 RMSE that is generally close to theensemble mean or the ensemble median.

To assess the impact of model’s drift inherited from thediversity of spin-up strategies (Fig. 1 and Table 1) on the per-formance metrics, we use a simple additive assumption toincorporate an incremental error due to the drift, 1RMSE,to the above-mentioned RMSE. This incremental error dueto the drift is computed using the relaxation time τ deter-mined from the piControl simulations of each CMIP5 modelat each depth level (Eq. 1 and Fig. 9) and a common durationof T = 3000 years for all models (m):

1RMSEm(z)=

T∫0

driftm(z, t = 0)× exp(−1τ(z)

t)dt (2)

where 1RMSE has the same unit as RMSE.The common duration T is used to bring model drift close

to zero and hence to make models comparable to each other.

We employ 1RMSE to penalize the distance from the ob-servations assuming that this drift-induced deviation in tracerfields can be added to RMSE. This means that the effectof the penalty is to increase the distance giving a consistentmeasure of the equilibration error.

The right-hand side panels of Fig. 10 show the influence ofthis penalization approach on the model ranking at the var-ious depth levels. They show that several models have beenupgraded in the ranking while others have not. For exam-ple, both MPI-ESM-LR, MPI-ESM-MR have been upgradedat the surface and 2000 m. On the other hand, the rank ofHadGEM2-ES and HadGEM2-CC has been downgraded tothe 5th and 3rd position due to the large drift in surface oxy-gen concentrations in comparison to that of the other mod-els. The surface drift might be attributed to drivers in oxy-gen fluxes (e.g., SST, SSS). The ranking of GFDL-ESM2Gand GFDL-ESM2M slightly changes with penalization butboth models stay close to the ensemble mean or the ensem-ble median. At the bottom of the ranking, models with largedeviation from the oxygen observations (i.e., CMCC-CESM,IPSL-CM5B-LR, NorESM1-ME, CNRM-CM5) are found.For these models, the computed 1RMSE and RMSE resultin similar ranking, because even a small drift and hence rela-tively low1RMSE cannot compensate for their large RMSE.

4 Discussion

4.1 Implications for biogeochemical processes

Our results show that errors in ocean biogeochemical fieldsamplify during the spin-up simulation but not at the same rateat all depths. These differences in error evolution are consis-tent with an increasing contribution of biogeochemical pro-cesses in setting the distribution of tracers at depth. Indeed,Mignot et al. (2013) with the same model simulation showedthat the main physical climate fields as well as the large-scaleocean circulation reach quasi-equilibrium after 250 years ofspin-up, but our analyses indicate that biogeochemical trac-ers do not (Fig. 3).

Besides, our analysis demonstrates that drifts in biogeo-chemical fields are highly model dependent. For example,despite having the same initialization strategy and compa-rable spin-up duration, the GFDL-ESM2G, GFDL-ESM2M,and NorESM1-ME models display considerable difference indrift (Figs. 9 and 10) that mirror large differences in modelperformance and properties (e.g., resolution, simulated pro-cesses).

The identification of the dynamical or biogeochemicalprocesses responsible for these errors is not within the scopeof this study and would require (or “would have required”)additional long simulations with additional tracers targetedfor attribution of the various biogeochemical processes andthe underlying ocean physics (e.g., Doney et al., 2004) in-volved (e.g. using abiotic, passive tracers as suggested in



0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32

#15 CMCC−CESM#14 IPSL−CM5B−LR#13 IPSL−CM5A−LR

#12 GFDL−ESM2M#11 CNRM−CM5

#10 CNRM−CM5−2#9 GFDL−ESM2G#8 CNRM−ESM1

#7 IPSL−CM5A−MR#6 NorESM1−ME#5 MPI−ESM−LR#4 CESM1−BGC

#3 MPI−ESM−MR#2 HadGEM2−CC#1 HadGEM2−ES (a)

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32

#14 CMCC−CESM#15 IPSL−CM5B−LR#8 IPSL−CM5A−LR#10 GFDL−ESM2M

#13 CNRM−CM5#6 CNRM−CM5−2

#11 GFDL−ESM2G#9 CNRM−ESM1

#12 IPSL−CM5A−MR#7 NorESM1−ME#1 MPI−ESM−LR#4 CESM1−BGC

#2 MPI−ESM−MR#3 HadGEM2−CC#5 HadGEM2−ES (d)

0 5 15 25 35 45 55 65 75 85 95 105 115

#15 IPSL−CM5B−LR#14 NorESM1−ME#13 CMCC−CESM#12 HadGEM2−CC#11 HadGEM2−ES

#10 IPSL−CM5A−LR#9 CESM1−BGC

#8 IPSL−CM5A−MR#7 GFDL−ESM2G

#6 CNRM−CM5#5 CNRM−CM5−2#4 GFDL−ESM2M#3 MPI−ESM−MR#2 MPI−ESM−LR#1 CNRM−ESM1 (b)

0 5 15 25 35 45 55 65 75 85 95 105 115

#13 IPSL−CM5B−LR#9 NorESM1−ME

#15 CMCC−CESM#8 HadGEM2−CC

#14 HadGEM2−ES#10 IPSL−CM5A−LR

#4 CESM1−BGC#12 IPSL−CM5A−MR

#7 GFDL−ESM2G#5 CNRM−CM5

#2 CNRM−CM5−2#1 GFDL−ESM2M#3 MPI−ESM−MR#6 MPI−ESM−LR

#11 CNRM−ESM1 (e)

Distance from observations for O2

0 20 40 60 80 100 120 140 160 180 200 220

#15 NorESM1−ME#14 CMCC−CESM

#13 IPSL−CM5B−LR#12 CNRM−CM5

#11 CNRM−CM5−2#10 GFDL−ESM2G

#9 CNRM−ESM1#8 MPI−ESM−LR

#7 IPSL−CM5A−MR#6 GFDL−ESM2M#5 MPI−ESM−MR#4 CESM1−BGC

#3 IPSL−CM5A−LR#2 HadGEM2−ES#1 HadGEM2−CC (c)

Distance from observations for O2 − penalized with drift0 20 40 60 80 100 120 140 160 180 200 220

#14 NorESM1−ME#15 CMCC−CESM

#12 IPSL−CM5B−LR#13 CNRM−CM5

#7 CNRM−CM5−2#6 GFDL−ESM2G#11 CNRM−ESM1#3 MPI−ESM−LR

#10 IPSL−CM5A−MR#9 GFDL−ESM2M#2 MPI−ESM−MR#5 CESM1−BGC

#8 IPSL−CM5A−LR#4 HadGEM2−ES#1 HadGEM2−CC (f)

(μmol L )–1 (μmol L )–1

Figure 10. Rankings of CMIP5 Earth system models based on the standard and penalized version of the distance from oxygen observations.The standard distance metric is calculated as the ensemble-mean root-mean squared error (RMSE) for O2 concentrations at surface (a),150 m (b) and 2000 m (c). The penalized distance metric incorporates drift-induced changes in O2 RMSE (1RMSE) into O2 RMSE atsurface (d), 150 m (e) and 2000 m (f). Ensemble-mean RMSEs are calculated using available ensemble members of Earth system modeloxygen concentrations averaged over the 1986–2005 historical period relative to WOA2013 observations. 1RMSE is determined usingEq. (2) and fits derived from the first century of the CMIP5 piControl simulations. Solid red and magenta lines indicate the multi-modelmean standard and penalized distance from O2 observations, respectively. With the same color pattern, dashed lines are indicative of themulti-model median for the standard and penalized distance from O2 observations.

Walin et al., 2014). Some mechanisms can be nonethelessinvoked to explain differences or similarities in behavior be-tween biogeochemical fields. For example, the evolution ofsurface concentrations for O2 and Alk-DIC is controlled inpart by the solubility of O2 and CO2 in seawater and theconcentration of these gases in the atmosphere (set to theobserved values and kept constant in all experiments per-formed with IPSL-CM5A-LR discussed here) and the bio-logical soft-tissue and calcium carbonate counter pumps (inrelation with the vertical transport of nutrients and alkalin-ity). Therefore, the equilibration of the O2 and Alk-DIC sur-face fields once the physical equilibrium is to a large de-gree reached (∼ 250 years of spin-up) is expected (Figs. 3a, cand 7). Nevertheless, spatial errors could increase depending

on the physical state of the model (Fig. 4b and f). By con-trast, the evolution of NO3 concentration is predominantlydetermined by ocean circulation, biological processes, andto a lesser extent by external supplies from rivers and atmo-sphere. Below the surface, concentrations of O2, NO3, andAlk-DIC evolve in response to the combined effect of oceancirculation and biogeochemical processes. The combinationof dynamical and biogeochemical processes on the one hand,and the spin-up strategy on the other hand both shape themodeled distributions of large-scale biogeochemical tracers.

Consequences of the difficulty in achieving the correctequilibration procedure have important implications for bio-geochemical features that are defined by regional charac-teristics in tracer concentrations, such as high nutrient/low



chlorophyll regions, oxygen minimum zones and nutrient-to-light colimitation patterns. This point is illustrated by recentstudies focusing on future changes in phytoplankton produc-tivity (e.g., Vancoppenolle et al., 2013, and Laufkötter et al.,2015). Vancoppenolle and co-workers report a wide spreadof surface mean NO3 concentrations (1980–1999) in the Arc-tic with a range from 1.7 to 8.9 µmol L−1 across a subset of11 CMIP5 models. The spread in present-day NO3 concen-trations translates into a large model-to-model uncertainty infuture net primary production. Laufkötter et al. (2015) de-termined limitation terms of phytoplankton production for asubset of CMIP5 and MAREMIP (Marine Ecosystem ModelIntercomparison Project) models. The authors demonstratethat nutrient-to-light colimitation patterns differ in strength,location and type between models and arise from large dif-ferences in the simulated nutrient concentrations. AlthoughVancoppenolle et al. (2013) and Laufkötter et al. (2015) ex-plain a part of the difference in simulated nutrient concen-tration by the differences in the spatial resolution and thecomplexity of the models, the authors of both studies qualita-tively invoked differences in spin-up duration to explain theremaining differences in simulated concentrations. Besides,a recent assessment of interannual to decadal variability ofocean–atmosphere CO2 and O2 fluxes in CMIP5 models,suggests that decadal variability can range regionally from10 to 50 % of the total natural variability among a subset of6 ESMs (Resplandy et al., 2015). In that study, the authorsdemonstrate that, despite the robustness of driving mecha-nisms (mostly related to vertical transport of water masses)across the model ensemble, model-to-model spread can berelated to differences in modeled carbon and oxygen con-centrations. In light of present results, it appears likely thatdifferences in spin-up strategy and sources of initializationcould also contribute to the amplitude of the natural variabil-ity of the ocean CO2 and O2 fluxes.

4.2 Implications for future projections

The inconsistent strategies used to spin up models in CMIP5have resulted in a significant source of uncertainty to themulti-model spread. It needs to be better constrained in orderto draw robust conclusions on the impact of climate changeon the carbon cycle as well as on climate feedback (e.g.,Arora et al., 2013; Friedlingstein et al., 2013; Roy et al.,2011; Schwinger et al., 2014; Séférian et al., 2012) and onmarine ecosystems (e.g., Bopp et al., 2013; Boyd et al., 2015;Cheung et al., 2012; Doney et al., 2012; Gattuso et al., 2015;Lehodey et al., 2006). So far, the most frequently used ap-proach relies on long preindustrial control simulations run-ning parallel to a transient simulations, allowing the “re-moval” of the drift in the simulated fields over the historicalperiod or future projections (e.g., Bopp et al., 2013; Coccoet al., 2013; Friedlingstein et al., 2013, 2006; Frölicher etal., 2014; Gehlen et al., 2014; Keller et al., 2014; Steinacheret al., 2010; Tjiputra et al., 2014). Although this approach

allows one to determine relative changes, it does not allowone to investigate the underlying reasons for the spread be-tween models in terms of processes, variability and responseto climate change. The “drift-correction” approach, much asthe one used for this study, assumes that drift-induced er-rors in the simulated fields can be isolated from the signal ofinterest. Verification of this fundamental hypothesis wouldrequire a specific experimental set-up consisting of the per-turbation of model fields (e.g., nutrients or carbon-relatedfields) to assess by how much the model projections wouldbe modified. So far, several modeling groups have gener-ated ensemble simulations in CMIP5 using a perturbationapproach. However, the perturbations were applied either tophysical fields only or to both the physical and marine bio-geochemical fields. To assess impacts of different spin-upstrategies and/or initial conditions on future projections ofmarine biogeochemical tracer distributions, ensemble sim-ulations in which only biogeochemical fields are perturbedwould be needed.

4.3 Implications for multi-model skill-scoreassessments

While the importance of spin-up protocols is well accepted inthe modeling community, the link between spin-up strategyand the ability of a model to reproduce modern observationsremains to be addressed.

Most of the recent CMIP5 skill assessment approacheswere based on historical hindcasts that were started frompreindustrial runs of varying duration and from various spin-up strategies. Therefore, in typical intercomparison exer-cises, Earth system models with a short spin-up, and hencemodeled distributions still close to initial fields, are con-fronted with Earth system models with a longer spin-up dura-tion and modeled distributions that have drifted further awayfrom their initial states. Our study highlights that such incon-sistencies in spin-up protocols and initial conditions acrossCMIP5 Earth system models (Fig. 1 and Table 1) couldsignificantly contribute to model-to-model spread in perfor-mance metrics. The analysis of the first century of CMIP5piControl simulations demonstrated a significant spread ofdrift between CMIP5 models (Fig. 9). An approximate ex-ponential relationship between the amplitude of the drift andthe spin-up duration emerges from the ensemble of CMIP5models, which is consistent with results from IPSL-CM5A-LR. For example, while the global average root-mean squareerror increased up to 70 % during a 500-year spin-up sim-ulation with IPSL-CM5A-LR, its rate of increase (or drift)decreased with time to a very small rate (0.001 Pg C y−2.Combining a simple drift model and this relationship, wepropose a penalization approach in an effort to assess moreobjectively the influence of documented model differenceson model–data biases. Figure 10 compares the standard ap-proach to assessing model performance (left-hand side pan-els) to the drift-penalized approach (right-hand side panels).



This novel approach penalizes models with larger drift with-out affecting the models with smaller drift. Taking into ac-count drift in modeled fields results in subtle adjustments inranking, which reflect differences in spin-up and initializa-tion strategies.

4.4 Limitations of the framework

In this work, the analyses focus on the globally averaged O2RMSE across a diverse ensemble of CMIP5 models, whichdiffer in terms of represented processes, spatial resolutionand performance in addition to differences in spin-up pro-tocols. Major limitations of the framework are presented be-low.

Due to their specificities in terms of processes and resolu-tion (e.g., Cabré et al., 2015; Laufkötter et al., 2015), regionaldrift in CMIP5 models may differ from the drift computedfrom globally averaged skill-score metrics (see Figs. S2 andS3). These differences may lead to different estimates of therelaxation time τ at a regional scale. Moreover, the combina-tion of regional ocean physics and biogeochemical processesin each individual model may drive an evolution of a regionaldrift in RMSE that does not fit the hypothesis of an exponen-tial decay of the drift during the course of the spin-up simu-lation.

Besides, a difference in the simulated processes and reso-lution in the different models can explain the relatively lowconfidence level of the fit to drift across the multi-modelCMIP5 ensemble (Fig. 9). The relatively low significancelevel of the fit reflects not only the large diversity of spin-upprotocols and initial conditions (Fig. 1 and Table 1), but alsothe large diversity of processes and resolution of the CMIP5models. Indeed, as shown in Kriest and Oschlies (2015), var-ious parameterizations of the particle sinking speed in a com-mon physical framework may lead to a similar evolution ofthe globally averaged RMSE in the first century of the spin-up simulation but display very different behavior within atimescale of O(103) years. As such, drift and τ estimatesneed to be used with caution when computed from short spin-up simulations because they can be subject to large uncer-tainties. An improved derivation of the penalization wouldrequire access to output from spin-up simulations for eachindividual model or, at least, a better quantification of model–model differences in terms of initial conditions.

5 Conclusions and recommendation for futureintercomparison exercises

Skill-score metrics are expected to be widely used in theframework of the upcoming CMIP6 (Meehl et al., 2014)with the development of international community bench-marking tools like the ESMValTool (http://www.pa.op.dlr.de/ESMValTool; see also Eyring et al., 2015). The assessment ofmodel skill to reproduce observations will focus on the mod-

ern period. Complementary to this approach, our results callfor the consideration of spin-up and initialization strategies inthe determination of skill assessment metrics (e.g., Friedrichset al., 2009; Stow et al., 2009) and, by extension, in modelweighting (e.g., Steinacher et al., 2010) and model ranking(e.g., Anav et al., 2013). Indeed, the use of equilibrium-statemetrics of the model like the three-dimensional drift of rel-evant skill-score metrics (e.g., RMSE) could be employedto increase the reliability of these traditional metrics and, assuch, should be included in the set of standard assessmenttools for CMIP6.

In an effort to better represent interactions between ma-rine biogeochemistry and climate (Smith et al., 2014), fu-ture generations of Earth system models are likely to includemore complex ocean biogeochemical models, be it in termsof processes (e.g., Tagliabue and Völker, 2011; Tagliabue etal., 2011) or interactions with other biogeochemical cycles(e.g., Gruber and Galloway, 2008) or increased spatial reso-lution (e.g., Dufour et al., 2013; Lévy et al., 2012) in order tobetter represent mesoscale biogeochemical dynamics. Thesedevelopments will go along with an increase in the diversityand complexity of spin-up protocols applied to Earth systemmodels, especially those including an interactive atmosphericCO2 or interactive nitrogen cycle (e.g., Dunne et al., 2013;Lindsay et al., 2014). The additional challenge of spinning-up emission-driven simulations with interactive carbon cyclewill also require us to extend the assessment of the impactof spin-up protocols to the terrestrial carbon cycle. Processessuch as soil carbon accumulation, peat formation as well asshift in biomes such as tropical and boreal ecosystems for dy-namic vegetation models require several long timescales toequilibrate (Brovkin et al., 2010; Koven et al., 2015). In ad-dition, the terrestrial carbon cycle has large uncertainties interms of carbon sink/source behavior (Anav et al., 2013; Dal-monech et al., 2014; Friedlingstein et al., 2013), which mightaffect ocean CO2 uptake (Brovkin et al., 2010). A novel nu-merical algorithm to accelerate the spin-up integration timefor computationally expensive ocean biogeochemical mod-els has emerged (Khatiwala, 2008), which could help to dis-entangle physical from biogeochemical contributions to theinter-model spread, but at the same time, could also poten-tially complicate the determination of inter-model spread byincreasing the diversity of spin-up protocols.

To evaluate the contribution of variable spin-up and ini-tialization strategies to model performance, these should bedocumented extensively and the corresponding model out-put should be archived. Ideally, for future coupled model in-tercomparison exercises (i.e., CMIP6, CMIP7, Meehl et al.,2014), the community should agree on a set of simple rec-ommendations for spin-up protocols, following past projectssuch as OCMIP-2. In parallel, any trade-off between modelequilibration and computationally efficient spin-up proce-dures has to be linked with efforts to reduce model errorsdue to the physical and biogeochemical parameterizations.


http://www.pa.op.dlr.de/ESMValTool

http://www.pa.op.dlr.de/ESMValTool


Supplement. The supplement related to this article is available on-line at: https://doi.org/10.5194/gmd-9-1827-2016-supplement.

Acknowledgements. We sincerely thank I. Kriest, F. Joos, theanonymous reviewer and A. Yool for their useful comments on thispaper. This work was supported by H2020 project CRESCENDO“Coordinated Research in Earth Systems and Climate: Experi-ments, kNowledge, Dissemination and Outreach”, which receivedfunding from the European Union’s Horizon 2020 research andinnovation programme under grant agreement no. 641816 and bythe EU FP7 project CARBOCHANGE “Changes in carbon uptakeand emissions by oceans in a changing climate” which receivedfunding from the European community’s Seventh Framework Pro-gramme under grant agreement no. 264879. Supercomputing timewas provided by GENCI (Grand Equipement National de CalculIntensif) at CCRT (Centre de Calcul Recherche et Technologie),allocation 016178. Finally, we are grateful to the ESGF projectwhich makes data available for all the community. Roland Séférianis grateful to Aurélien Ribes for his kind advices on statistics.Jerry Tjiputra acknowledges ORGANIC project (239965/F20)funded by the Research Council of Norway. Christoph Heinzeand Jerry Tjiputra are grateful for support through project EVA –Earth system modelling of climate variations in the Anthropocene(229771/E10) funded by the Research Council of Norway, as wellas CPU-time and mass storage provided through NOTUR projectNN2345K as well as NorStore project NS2345K. Keith Lindsayand Scott C. Doney acknowledge support from the NationalScience Foundation.

Edited by: A. Yool

References

Adachi, Y., Yukimoto, S., Deushi, M., Obata, A., Nakano,H., Tanaka, T. Y., Hosaka, M., Sakami, T., Yoshimura, H.,Hirabara, M., Shindo, E., Tsujino, H., Mizuta, R., Yabu, S.,Koshiro, T., Ose, T., and Kitoh, A.: Basic performance ofa new earth system model of the Meteorological ResearchInstitute (MRI-ESM1), Pap. Meteorol. Geophys., 64, 1–18,https://doi.org/10.2467/mripapers.64.1, 2013.

Anav, A., Friedlingstein, P., Kidston, M., Bopp, L., Ciais, P., Cox,P., Jones, C., Jung, M., Myneni, R., and Zhu, Z.: Evaluating theLand and Ocean Components of the Global Carbon Cycle inthe CMIP5 Earth System Models, J. Climate, 26, 6801–6843,https://doi.org/10.1175/JCLI-D-12-00417.1, 2013.

Andrews, O. D., Bindoff, N. L., Halloran, P. R., Ilyina, T., and LeQuéré, C.: Detecting an external influence on recent changesin oceanic oxygen using an optimal fingerprinting method,Biogeosciences, 10, 1799–1813, https://doi.org/10.5194/bg-10-1799-2013, 2013.

Archer, D., Buffett, B., and Brovkin, V.: Ocean methane hydratesas a slow tipping point in the global carbon cycle, P. Natl. Acad.Sci., 106, 20596–20601, 2009.

Arora, V. K., Scinocca, J. F., Boer, G. J., Christian, J. R., Denman,K. L., Flato, G. M., Kharin, V. V., Lee, W. G., and Merryfield,W. J.: Carbon emission limits required to satisfy future repre-sentative concentration pathways of greenhouse gases, Geophys.

Res. Lett., 38, L05805, https://doi.org/10.1029/2010GL046270,2011.

Arora, V. K., Boer, G. J., Friedlingstein, P., Eby, M., Jones, C.D., Christian, J. R., Bonan, G., Bopp, L., Brovkin, V., Cad-ule, P., Hajima, T., Ilyina, T., Lindsay, K., Tjiputra, J. F., andWu, T.: Carbon–Concentration and Carbon–Climate Feedbacksin CMIP5 Earth System Models, J. Climate, 26, 5289–5314,https://doi.org/10.1175/JCLI-D-12-00494.1, 2013.

Aumont, O. and Bopp, L.: Globalizing results from ocean in situiron fertilization studies, Global Biogeochem. Cy., 20, GB2017,https://doi.org/10.1029/2005GB002591, 2006.

Aumont, O., Orr, J., Jamous, D., Monfray, P., Marti, O., and Madec,G.: A degradation approach to accelerate simulations to steady-state in a 3-D tracer transport model of the global ocean, Clim.Dynam., 14, 101–116, 1998.

Aumont, O., Orr, J. C., Monfray, P., Ludwig, W., Amiotte-Suchet, P., and Probst, J.-L.: Riverine-driven interhemispherictransport of carbon, Global Biogeochem. Cy., 15, 393–405,https://doi.org/10.1029/1999GB001238, 2001.

Aumont, O., Maier-Reimer, E., Blain, S., and Monfray, P.:An ecosystem model of the global ocean including Fe,Si, P colimitations, Global Biogeochem. Cy., 17, 1060,https://doi.org/10.1029/2001GB001745, 2003.

Bopp, L., Resplandy, L., Orr, J. C., Doney, S. C., Dunne, J. P.,Gehlen, M., Halloran, P., Heinze, C., Ilyina, T., Séférian, R.,Tjiputra, J., and Vichi, M.: Multiple stressors of ocean ecosys-tems in the 21st century: projections with CMIP5 models,Biogeosciences, 10, 6225–6245, https://doi.org/10.5194/bg-10-6225-2013, 2013.

Boyd, P. W., Lennartz, S. T., Glover, D. M., and Doney, S. C.: Bio-logical ramifications of climate-change-mediated oceanic multi-stressors, Nature Clim. Change, 5, 71–79, 2015.

Bretherton, C. S., Widmann, M., Dymnikov, V. P., Wal-lace, J. M., and Bladé, I.: The Effective Number ofSpatial Degrees of Freedom of a Time-Varying Field,J. Climate, 12, 1990–2009, https://doi.org/10.1175/1520-0442(1999)012<1990:TENOSD>2.0.CO;2, 1999.

Brovkin, V., Lorenz, S. J., Jungclaus, J., Raddatz, T., Timm-reck, C., Reick, C. H., Segschneider, J., and Six, K.: Sensitiv-ity of a coupled climate-carbon cycle model to large volcaniceruptions during the last millennium, Tellus B, 62, 674–681,https://doi.org/10.1111/j.1600-0889.2010.00471.x, 2010.

Bryan, K.: Accelerating the Convergence to Equi-librium of Ocean-Climate Models, J. Phys.Oceanogr., 14, 666–673, https://doi.org/10.1175/1520-0485(1984)014<0666:ATCTEO>2.0.CO;2, 1984.

Cabré, A., Marinov, I., Bernardello, R., and Bianchi, D.: Oxy-gen minimum zones in the tropical Pacific across CMIP5 mod-els: mean state differences and climate change trends, Biogeo-sciences, 12, 5429–5454, https://doi.org/10.5194/bg-12-5429-2015, 2015.

Cheung, W. W. L., Sarmiento, J. L., Dunne, J. P., Frölicher, T. L.,Lam, V. W. Y., Palomares, M. L. D., Watson, R., and Pauly,D.: Shrinking of fishes exacerbates impacts of global oceanchanges on marine ecosystems, Nature Climate Change, 2, 1–5,https://doi.org/10.1038/nclimate1691, 2012.

Cocco, V., Joos, F., Steinacher, M., Frölicher, T. L., Bopp,L., Dunne, J., Gehlen, M., Heinze, C., Orr, J., Oschlies,A., Schneider, B., Segschneider, J., and Tjiputra, J.: Oxy-


https://doi.org/10.5194/gmd-9-1827-2016-supplement

https://doi.org/10.2467/mripapers.64.1

https://doi.org/10.1175/JCLI-D-12-00417.1

https://doi.org/10.5194/bg-10-1799-2013

https://doi.org/10.5194/bg-10-1799-2013

https://doi.org/10.1029/2010GL046270


https://doi.org/10.1029/2005GB002591

https://doi.org/10.1029/1999GB001238

https://doi.org/10.1029/2001GB001745

https://doi.org/10.5194/bg-10-6225-2013

https://doi.org/10.5194/bg-10-6225-2013

https://doi.org/10.1175/1520-0442(1999)012<1990:TENOSD>2.0.CO;2

https://doi.org/10.1175/1520-0442(1999)012<1990:TENOSD>2.0.CO;2

https://doi.org/10.1111/j.1600-0889.2010.00471.x

https://doi.org/10.1175/1520-0485(1984)014<0666:ATCTEO>2.0.CO;2

https://doi.org/10.1175/1520-0485(1984)014<0666:ATCTEO>2.0.CO;2

https://doi.org/10.5194/bg-12-5429-2015

https://doi.org/10.5194/bg-12-5429-2015

https://doi.org/10.1038/nclimate1691


gen and indicators of stress for marine life in multi-modelglobal warming projections, Biogeosciences, 10, 1849–1868,https://doi.org/10.5194/bg-10-1849-2013, 2013.

Collins, W. J., Bellouin, N., Doutriaux-Boucher, M., Gedney, N.,Halloran, P., Hinton, T., Hughes, J., Jones, C. D., Joshi, M., Lid-dicoat, S., Martin, G., O’Connor, F., Rae, J., Senior, C., Sitch,S., Totterdell, I., Wiltshire, A., and Woodward, S.: Developmentand evaluation of an Earth-System model – HadGEM2, Geosci.Model Dev., 4, 1051–1075, https://doi.org/10.5194/gmd-4-1051-2011, 2011.

Cox, P. M., Pearson, D., Booth, B. B., Friedlingstein, P., Hunting-ford, C., Jones, C. D., and Luke, C. M.: Sensitivity of tropicalcarbon to climate change constrained by carbon dioxide variabil-ity, Nature, 494, 341–344, https://doi.org/10.1038/nature11882,2013.

Dalmonech, D., Foley, A. M., Anav, A., Friedlingstein, P., Friend,A. D., Kidston, M., Willeit, M., and Zaehle, S.: Challengesand opportunities to reduce uncertainty in projections of fu-ture atmospheric CO2: a combined marine and terrestrial bio-sphere perspective, Biogeosciences Discuss., 11, 2083–2153,https://doi.org/10.5194/bgd-11-2083-2014, 2014.

de Baar, H. J. W. and de Jong, J. T. M.: The biogeochemistry ofiron in seawater, edited by: Turner, D. R. and Hunter, K. A., JohnWiley, Hoboken, NJ, USA, 2001.

Doney, S. C.: The Growing Human Footprint on Coastaland Open-Ocean Biogeochemistry, Science, 328, 1512–1516,https://doi.org/10.1126/science.1185198, 2010.

Doney, S. C., Lindsay, K., Caldeira, K., Campin, J.-M., Drange,H., Dutay, J.-C., Follows, M., Gao, Y., Gnanadesikan, A., Gru-ber, N., Ishida, A., Joos, F., Madec, G., Maier-Reimer, E.,Marshall, J. C., Matear, R. J., Monfray, P., Mouchet, A., Naj-jar, R., Orr, J. C., Plattner, G.-K., Sarmiento, J., Schlitzer, R.,Slater, R., Totterdell, I. J., Weirig, M.-F., Yamanaka, Y., andYool, A: Evaluating global ocean carbon models: The impor-tance of realistic physics, Global Biogeochem. Cy., 18, GB3017,https://doi.org/10.1029/2003GB002150, 2004.

Doney, S. C., Lima, I., Moore, J. K., Lindsay, K., Behren-feld, M. J., Westberry, T. K., Mahowald, N., Glover, D.M., and Takahashi, T.: Skill metrics for confronting globalupper ocean ecosystem-biogeochemistry models against fieldand remote sensing data, J. Marine Syst., 76, 95–112,https://doi.org/10.1016/j.jmarsys.2008.05.015, 2009.

Doney, S. C., Ruckelshaus, M., Emmett Duffy, J., Barry, J. P.,Chan, F., English, C. A., Galindo, H. M., Grebmeier, J. M.,Hollowed, A. B., Knowlton, N., Polovina, J., Rabalais, N. N.,Sydeman, W. J., and Talley, L. D.: Climate Change Impactson Marine Ecosystems, Annu. Rev. Marine. Sci., 4, 11–37,https://doi.org/10.1146/annurev-marine-041911-111611, 2012.

Dufour, C. O., Sommer, J. L., Gehlen, M., Orr, J. C., Molines, J.-M., Simeon, J., and Barnier, B.: Eddy compensation and controlsof the enhanced sea-to-air CO2 flux during positive phases of theSouthern Annular Mode, Global Biogeochem. Cy., 27, 950–961,https://doi.org/10.1002/gbc.20090, 2013.

Dufresne, J.-L., Foujols, M. A., Denvil, S., Caubel, A., Marti, O.,Aumont, O., Balkanski, Y., Bekki, S., Bellenger, H., Benshila,R., Bony, S., Bopp, L., Braconnot, P., Brockmann, P., Cadule, P.,Cheruy, F., Codron, F., Cozic, A., Cugnet, D., Noblet, N., Duvel,J. P., Ethe, C., Fairhead, L., Fichefet, T., Flavoni, S., Friedling-stein, P., Grandpeix, J. Y., Guez, L., Guilyardi, E., Hauglus-

taine, D., Hourdin, F., Idelkadi, A., Ghattas, J., Joussaume, S.,Kageyama, M., Krinner, G., Labetoulle, S., Lahellec, A., Lefeb-vre, M.-P., Lefèvre, F., Lévy, C., Li, Z. X., Lloyd, J., Lott, F.,Madec, G., Mancip, M., Marchand, M., Masson, S., Meurdes-oif, Y., Mignot, J., Musat, I., Parouty, S., Polcher, J., Rio, C.,Schulz, M., Swingedouw, D., Szopa, S., Talandier, C., Terray, P.,Viovy, N., and Vuichard, N.: Climate change projections usingthe IPSL-CM5 Earth System Model: from CMIP3 to CMIP5,Clim. Dynam., 40, 2123–2165, https://doi.org/10.1007/s00382-012-1636-1, 2013.

Dunne, J. P., John, J. G., Adcroft, A. J., Griffies, S. M., Hallberg,R. W., Shevliakova, E., Stouffer, R. J., Cooke, W., Dunne, K. A.,Harrison, M. J., Krasting, J. P., Malyshev, S. L., Milly, P. C. D.,Phillipps, P. J., Sentman, L. A., Samuels, B. L., Spelman, M. J.,Winton, M., Wittenberg, A. T., and Zadeh, N.: GFDL’s ESM2Global Coupled Climate–Carbon Earth System Models. Part I:Physical Formulation and Baseline Simulation Characteristics,J. Climate, 25, 6646–6665, https://doi.org/10.1175/JCLI-D-11-00560.1, 2013.

Duplessy, J. C., Bard, E., Arnold, M., Shackleton, N. J., Duprat, J.,and Labeyrie, L.: How fast did the ocean—atmosphere systemrun during the last deglaciation?, Earth Planet. Sc. Lett., 103, 27–40, https://doi.org/10.1016/0012-821X(91)90147-A, 1991.

Eyring, V., Righi, M., Evaldsson, M., Lauer, A., Wenzel, S., Jones,C., Anav, A., Andrews, O., Cionni, I., Davin, E. L., Deser, C.,Ehbrecht, C., Friedlingstein, P., Gleckler, P., Gottschaldt, K.-D., Hagemann, S., Juckes, M., Kindermann, S., Krasting, J.,Kunert, D., Levine, R., Loew, A., Mäkelä, J., Martin, G., Ma-son, E., Phillips, A., Read, S., Rio, C., Roehrig, R., Senftleben,D., Sterl, A., van Ulft, L. H., Walton, J., Wang, S., and Williams,K. D.: ESMValTool (v1.0) – a community diagnostic and per-formance metrics tool for routine evaluation of Earth SystemModels in CMIP, Geosci. Model Dev. Discuss., 8, 7541–7661,https://doi.org/10.5194/gmdd-8-7541-2015, 2015.

Fichefet, T. and Maqueda, M. A. M.: Sensitivity of a global sea icemodel to the treatment of ice thermodynamics and dynamics, J.Geophys. Res., 102, 12609–12646, 1997.

Follows, M. J., Dutkiewicz, S., Grant, S., and Chisholm,S. W.: Emergent Biogeography of Microbial Commu-nities in a Model Ocean, Science, 315, 1843–1846,https://doi.org/10.1126/science.1138544, 2007.

Friedlingstein, P., Cox, P., Betts, R., Bopp, L., Bloh, Von, W.,Brovkin, V., Cadule, P., Doney, S., Eby, M., Fung, I., Bala,G., John, J., Jones, C., Joos, F., Kato, T., Kawamiya, M.,Knorr, W., Lindsay, K., Matthews, H. D., Raddatz, T., Rayner,P., Reick, C., Roeckner, E., Schnitzler, K. G., Schnur, R.,Strassmann, K., Weaver, A. J., Yoshikawa, C., and Zeng, N.:Climate–Carbon Cycle Feedback Analysis: Results from theC4MIP Model Intercomparison, J. Climate, 10, 3337–3353,https://doi.org/10.1175/JCLI3800.1, 2006.

Friedlingstein, P., Meinshausen, M., Arora, V. K., Jones, C. D.,Anav, A., Liddicoat, S. K., and Knutti, R.: Uncertainties inCMIP5 climate projections due to carbon cycle feedbacks, J. Cli-mate, 27, 511–526, https://doi.org/10.1175/JCLI-D-12-00579.1,2013.

Friedrichs, M. A. M., Dusenberry, J. A., Anderson, L. A., Arm-strong, R. A., Chai, F., Christian, J. R., Doney, S. C., Dunne,J. P., Fujii, M., Hood, R., McGillicuddy Jr., D. J., Moore, J. K.,Schartau, M., Spitz, Y. H., and Wiggert, J. D.: Assessment of skill


https://doi.org/10.5194/bg-10-1849-2013

https://doi.org/10.5194/gmd-4-1051-2011

https://doi.org/10.5194/gmd-4-1051-2011

https://doi.org/10.1038/nature11882

https://doi.org/10.5194/bgd-11-2083-2014

https://doi.org/10.1126/science.1185198

https://doi.org/10.1029/2003GB002150

https://doi.org/10.1016/j.jmarsys.2008.05.015

https://doi.org/10.1146/annurev-marine-041911-111611

https://doi.org/10.1002/gbc.20090

https://doi.org/10.1007/s00382-012-1636-1

https://doi.org/10.1007/s00382-012-1636-1



https://doi.org/10.1016/0012-821X(91)90147-A

https://doi.org/10.5194/gmdd-8-7541-2015


https://doi.org/10.1175/JCLI3800.1



and portability in regional marine biogeochemical models: Roleof multiple planktonic groups, J. Geophys. Res., 112, C08001,https://doi.org/10.1029/2006JC003852, 2007.

Friedrichs, M. A. M., Carr, M.-E., Barber, R. T., Scardi, M.,Antoine, D., Armstrong, R. A., Asanuma, I., Behrenfeld, M.J., Buitenhuis, E. T., Chai, F., Christian, J. R., Ciotti, A. M.,Doney, S. C., Dowell, M., Dunne, J. P., Gentili, B., Gregg,W., Hoepffner, N., Ishizaka, J., Kameda, T., Lima, I., Marra,J., Mélin, F., Moore, J. K., Morel, A., O’Malley, R. T.,O’Reilly, J., Saba, V. S., Schmeltz, M., Smyth, T. J., Tjipu-tra, J., Waters, K., Westberry, T. K., and Winguth, A.: Assess-ing the uncertainties of model estimates of primary productiv-ity in the tropical Pacific Ocean, J. Marine Syst., 76, 113–133,https://doi.org/10.1016/j.jmarsys.2008.05.010, 2009.

Frölicher, T. L., Sarmiento, J. L., Paynter, D. J., Dunne, J. P., Krast-ing, J. P., and Winton, M.: Dominance of the Southern Ocean inanthropogenic carbon and heat uptake in CMIP5 models, J. Cli-mate, 28, 862–886, https://doi.org/10.1175/JCLI-D-14-00117.1,2014.

Gattuso, J. P., Magnan, A., Bille, R., Cheung, W. W. L., Howes,E. L., Joos, F., Allemand, D., Bopp, L., Cooley, S. R., Eakin, C.M., Hoegh-Guldberg, O., Kelly, R. P., Portner, H. O., Rogers,A. D., Baxter, J. M., Laffoley, D., Osborn, D., Rankovic,A., Rochette, J., Sumaila, U. R., Treyer, S., and Turley, C.:Contrasting futures for ocean and society from different an-thropogenic CO2 emissions scenarios, Science, 349, 6243,https://doi.org/10.1126/science.aac4722, 2015.

Gehlen, M., Séférian, R., Jones, D. O. B., Roy, T., Roth, R.,Barry, J., Bopp, L., Doney, S. C., Dunne, J. P., Heinze, C.,Joos, F., Orr, J. C., Resplandy, L., Segschneider, J., and Tjipu-tra, J.: Projected pH reductions by 2100 might put deep NorthAtlantic biodiversity at risk, Biogeosciences, 11, 6955–6967,https://doi.org/10.5194/bg-11-6955-2014, 2014.

Gerber, M. and Joos, F.: Carbon sources and sinks from an Ensem-ble Kalman Filter ocean data assimilation, Global Biogeochem.Cy., 24, GB3004, https://doi.org/10.1029/2009GB003531, 2010.

Gnanadesikan, A.: Oceanic ventilation and biogeochemical cycling:Understanding the physical mechanisms that produce realisticdistributions of tracers and productivity, Global Biogeochem.Cy., 18, GB4010, https://doi.org/10.1029/2003GB002097, 2004.

Gruber, N.: Warming up, turning sour, losing breath: ocean biogeo-chemistry under global change, Philos. T. R. Soc. A, 369, 1980–1996, https://doi.org/10.1098/rsta.2011.0003, 2011.

Gruber, N. and Galloway, J. N.: An Earth-system perspec-tive of the global nitrogen cycle, Nature, 451, 293–296,https://doi.org/10.1038/nature06592, 2008.

Hajima, T., Kawamiya, M., Watanabe, M., Kato, E., Tachiiri,K., Sugiyama, M., Watanabe, S., Okajima, H., and Ito, A.:Modeling in Earth system science up to and beyond IPCCAR5, Progress in Earth and Planetary Science, 1, 29–25,https://doi.org/10.1186/s40645-014-0029-y, 2014.

Heinze, C., Maier-Reimer, E., Winguth, A., and Archer, D.: Aglobal oceanic sediment model for long-term climate studies,Global Biogeochem. Cy., 13, 221–250, 1999.

Heinze, M. and Ilyina, T.: Ocean biogeochemistry in thewarm climate of the late Paleocene, Clim. Past, 11, 63–79,https://doi.org/10.5194/cp-11-63-2015, 2015.

Henson, S. A., Sarmiento, J. L., Dunne, J. P., Bopp, L.,Lima, I., Doney, S. C., John, J., and Beaulieu, C.: Detec-

tion of anthropogenic climate change in satellite records ofocean chlorophyll and productivity, Biogeosciences, 7, 621–640,https://doi.org/10.5194/bg-7-621-2010, 2010.

Hobbs, W., Palmer, M. D., and Monselesan, D.: An En-ergy Conservation Analysis of Ocean Drift in the CMIP5Global Coupled Models, J. Climate, 29, 1639–1653,https://doi.org/10.1175/JCLI-D-15-0477.1, 2015.

Hourdin, F., Musat, I., Bony, S., Braconnot, P., Codron, F.,Dufresne, J.-L., Fairhead, L., Filiberti, M.-A., Friedlingstein,P., Grandpeix, J.-Y., Krinner, G., LeVan, P., Li, Z.-X., andLott, F.: The LMDZ4 general circulation model: climate per-formance and sensitivity to parametrized physics with em-phasis on tropical convection, Clim. Dynam., 27, 787–813,https://doi.org/10.1007/s00382-006-0158-0, 2006.

Ilyina, T., Six, K. D., Segschneider, J., Maier-Reimer, E., Li,H., and Núñez-Riboni, I.: Global ocean biogeochemistry modelHAMOCC: Model architecture and performance as componentof the MPI-Earth system model in different CMIP5 experi-mental realizations, J. Adv. Model. Earth Syst., 5, 287–315,https://doi.org/10.1029/2012MS000178, 2013.

IPCC: Climate Change 2013: The Physical Science Basis, editedby: Stoker, T. F., Qin, D., Plattner, G., Tignor, M., Allen, S. K.,Boschung, J., Nauels, A., Xia, Y., Bex, V., and Midgley, P. M.,Cambridge Univ. Press, Cambridge, UK, and New York, NY,USA, 2013.

Ito, T. and Deutsch, C.: Variability of the Oxygen Mini-mum Zone in the Tropical North Pacific during the Late20th Century, Global Biogeochem. Cy., 27, 1119–1128,https://doi.org/10.1002/2013GB004567, 2013.

Ito, T., Woloszyn, M., and Mazloff, M.: Anthropogenic carbon diox-ide transport in the Southern Ocean driven by Ekman flow, Na-ture, 463, 80–83, https://doi.org/10.1038/nature08687, 2010.

Jickells, T. and Spokes, L.: The biogeochemistry of iron in seawater,edited by: Turner, D. R. and Hunter, K. A., John Wiley, Hoboken,NJ, USA, 2001.

Johnson, K., Chavez, F., and Friederich, G.: Continental-shelf sedi-ment as a primary source of iron for coastal phytoplankton, Na-ture, 398, 697–700, 1999.

Keeling, R. F., Körtzinger, A., and Gruber, N.: Ocean Deoxygena-tion in a Warming World, Annu. Rev. Marine. Sci., 2, 199–229,https://doi.org/10.1146/annurev.marine.010908.163855, 2009.

Keenlyside, N. S., Latif, M., Jungclaus, J., Kornblueh, L.,and Roeckner, E.: Advancing decadal-scale climate pre-diction in the North Atlantic sector, Nature, 453, 84–88,https://doi.org/10.1038/nature06921, 2008.

Keller, K. M., Joos, F., and Raible, C. C.: Time of emergenceof trends in ocean biogeochemistry, Biogeosciences, 11, 3647–3659, https://doi.org/10.5194/bg-11-3647-2014, 2014.

Key, R., Kozyr, A., Sabine, C., Lee, K., Wanninkhof, R., Bullis-ter, J., Feely, R., Millero, F., Mordy, C., and Peng, T.: A globalocean carbon climatology: Results from Global Data Analy-sis Project (GLODAP), Global Biogeochem. Cy., 18, GB4031,https://doi.org/10.1029/2004GB002247, 2004.

Khatiwala, S.: Fast spin up of ocean biogeochemical models usingmatrix-free Newton-Krylov, Ocean Model., 23, 121–129, 2008.

Khatiwala, S., Visbeck, M., and Cane, M. A.: Ac-celerated simulation of passive tracers in oceancirculation models, Ocean Model., 9, 51–69,https://doi.org/10.1016/j.ocemod.2004.04.002, 2005.


https://doi.org/10.1029/2006JC003852



https://doi.org/10.1126/science.aac4722

https://doi.org/10.5194/bg-11-6955-2014

https://doi.org/10.1029/2009GB003531

https://doi.org/10.1029/2003GB002097

https://doi.org/10.1098/rsta.2011.0003


https://doi.org/10.1186/s40645-014-0029-y

https://doi.org/10.5194/cp-11-63-2015

https://doi.org/10.5194/bg-7-621-2010


https://doi.org/10.1007/s00382-006-0158-0

https://doi.org/10.1029/2012MS000178

https://doi.org/10.1002/2013GB004567


https://doi.org/10.1146/annurev.marine.010908.163855


https://doi.org/10.5194/bg-11-3647-2014

https://doi.org/10.1029/2004GB002247

https://doi.org/10.1016/j.ocemod.2004.04.002


Kim, H.-M., Webster, P. J., and Curry, J. A.: Evaluationof short-term climate change prediction in multi-modelCMIP5 decadal hindcasts, Geophys. Res. Lett., 39, L10701,https://doi.org/10.1029/2012GL051644, 2012.

Knutti, R., Masson, D., and Gettelman, A.: Climate model geneal-ogy: Generation CMIP5 and how we got there, Geophys. Res.Lett., 40, 1194–1199, https://doi.org/10.1002/grl.50256, 2013.

Koven, C. D., Chambers, J. Q., Georgiou, K., Knox, R., Negron-Juarez, R., Riley, W. J., Arora, V. K., Brovkin, V., Friedlingstein,P., and Jones, C. D.: Controls on terrestrial carbon feedbacks byproductivity versus turnover in the CMIP5 Earth System Models,Biogeosciences, 12, 5211–5228, https://doi.org/10.5194/bg-12-5211-2015, 2015.

Kriest, I. and Oschlies, A.: MOPS-1.0: towards a model for theregulation of the global oceanic nitrogen budget by marinebiogeochemical processes, Geosci. Model Dev., 8, 2929–2957,https://doi.org/10.5194/gmd-8-2929-2015, 2015.

Krinner, G., Viovy, N., de Noblet-Ducoudré, N., Ogée, J., Polcher,J., Friedlingstein, P., Ciais, P., Sitch, S., and Prentice, I. C.:A dynamic global vegetation model for studies of the coupledatmosphere-biosphere system, Global Biogeochem. Cy., 19, 1–33, 2005.

Laufkötter, C., Vogt, M., Gruber, N., Aita-Noguchi, M., Aumont,O., Bopp, L., Buitenhuis, E., Doney, S. C., Dunne, J., Hashioka,T., Hauck, J., Hirata, T., John, J., Le Quéré, C., Lima, I. D.,Nakano, H., Seferian, R., Totterdell, I., Vichi, M., and Völker,C.: Drivers and uncertainties of future global marine primary pro-duction in marine ecosystem models, Biogeosciences, 12, 6955–6984, https://doi.org/10.5194/bg-12-6955-2015, 2015.

Lehodey, P., Alheit, J., Barange, M., Baumgartner, T., Beaugrand,G., Drinkwater, K., Fromentin, J.-M., Hare, S. R., Ottersen, G.,Perry, R. I., Roy, C., van der Lingen, C. D., and Werner, F.: Cli-mate variability, fish, and fisheries, J. Climate, 19, 5009–5030,https://doi.org/10.1175/JCLI3898.1, 2006.

Le Quéré, C., Moriarty, R., Andrew, R. M., Peters, G. P., Ciais, P.,Friedlingstein, P., Jones, S. D., Sitch, S., Tans, P., Arneth, A.,Boden, T. A., Bopp, L., Bozec, Y., Canadell, J. G., Chini, L. P.,Chevallier, F., Cosca, C. E., Harris, I., Hoppema, M., Houghton,R. A., House, J. I., Jain, A. K., Johannessen, T., Kato, E., Keel-ing, R. F., Kitidis, V., Klein Goldewijk, K., Koven, C., Landa,C. S., Landschützer, P., Lenton, A., Lima, I. D., Marland, G.,Mathis, J. T., Metzl, N., Nojiri, Y., Olsen, A., Ono, T., Peng, S.,Peters, W., Pfeil, B., Poulter, B., Raupach, M. R., Regnier, P., Rö-denbeck, C., Saito, S., Salisbury, J. E., Schuster, U., Schwinger,J., Séférian, R., Segschneider, J., Steinhoff, T., Stocker, B. D.,Sutton, A. J., Takahashi, T., Tilbrook, B., van der Werf, G. R.,Viovy, N., Wang, Y.-P., Wanninkhof, R., Wiltshire, A., and Zeng,N.: Global carbon budget 2014, Earth Syst. Sci. Data, 7, 47–85,https://doi.org/10.5194/essd-7-47-2015, 2015.

Levitus, S. and Boyer, T.: World ocean atlas 1994, volume 4: Tem-perature, PB–95-270112/XAB, National Environmental Satel-lite, Data, and Information Service, Washington, DC, USA,1994.

Levitus, S., Conkright, M. E., Reid, J. L., Najjar, R. G., andMantyla, A.: Distribution of nitrate, phosphate and silicate in theworld oceans, Prog. Oceanogr., 31, 245–273, 1993.

Levitus, S., Antonov, J. I., Baranova, O. K., Boyer, T. P., Cole-man, C. L., Garcia, H. E., Grodsky, A. I., Johnson, D. R., Lo-carnini, R. A., Mishonov, A. V., Reagan, J. R., Sazama, C. L.,

Seidov, D., Smolyar, I., Yarosh, E. S., and Zweng, M. M.: TheWorld Ocean Database TI, Data Science Journal, 12, WDS229–WDS234, 2013.

Lévy, M., Lengaigne, M., Bopp, L., Vincent, E. M., Madec,G., Ethe, C., Kumar, D., and Sarma, V. V. S. S.: Con-tribution of tropical cyclones to the air-sea CO2 flux:A global view, Global Biogeochem. Cy., 26, GB2001,https://doi.org/10.1029/2011GB004145, 2012.

Lindsay, K., Bonan, G. B., Doney, S. C., Hoffman, F. M.,Lawrence, D. M., Long, M. C., Mahowald, N. M., Moore, J.K., Randerson, J. T., and Thornton, P. E.: Preindustrial Con-trol and 20th Century Carbon Cycle Experiments with theEarth System Model CESM1(BGC), J. Climate, 27, 8981–9005,https://doi.org/10.1175/JCLI-D-12-00565.1, 2014.

Ludwig, W., Probst, J., and Kempe, S.: Predicting the oceanic inputof organic carbon by continental erosion, Global Biogeochem.Cy., 10, 23–41, 1996.

Madec, G.: NEMO ocean engine, Institut Pierre-Simon Laplace(IPSL), France, available at: http://www.nemo-ocean.eu/About-NEMO/Reference-manuals (last access: November2013), 2008.

Maier-Reimer, E.: Geochemical cycles in an ocean general cir-culation model. Preindustrial tracer distributions, Global Bio-geochem. Cy., 7, 645–677, https://doi.org/10.1029/93GB01355,1993.

Maier-Reimer, E. and Hasselmann, K.: Transport andstorage of CO2 in the ocean – an inorganic ocean-circulation carbon cycle model, Clim. Dynam., 2, 63–90,https://doi.org/10.1007/BF01054491, 1987.

Marinov, I., Gnanadesikan, A., Sarmiento, J. L., Toggweiler, J.R., Follows, M., and Mignone, B. K.: Impact of oceaniccirculation on biological carbon storage in the ocean andatmospheric pCO2, Global Biogeochem. Cy., 22, GB3007,https://doi.org/10.1029/2007GB002958, 2008.

Massonnet, F., Fichefet, T., Goosse, H., Bitz, C. M., Philippon-Berthier, G., Holland, M. M., and Barriat, P.-Y.: Constrainingprojections of summer Arctic sea ice, The Cryosphere, 6, 1383–1394, https://doi.org/10.5194/tc-6-1383-2012, 2012.

Matei, D., Baehr, J., Jungclaus, J. H., Haak, H., Muller, W. A.,and Marotzke, J.: Multiyear Prediction of Monthly Mean At-lantic Meridional Overturning Circulation at 26.5◦ N, Science,335, 76–79, https://doi.org/10.1126/science.1210299, 2012.

Mawji, E., Schlitzer, R., Dodas, E. M., et al.: The GEOTRACESIntermediate Data Product 2014, Marine Chemistry, 177, Part 1,1–8, https://doi.org/10.1016/j.marchem.2015.04.005.

Meehl, G. A., Goddard, L., Murphy, J., Stouffer, R. J.,Boer, G., Danabasoglu, G., Dixon, K., Giorgetta, M. A.,Greene, A. M., Hawkins, E., Hegerl, G., Karoly, D., Keenly-side, N., Kimoto, M., Kirtman, B., Navarra, A., Pul-warty, R., Smith, D., Stammer, D., and Stockdale, T.:Decadal Prediction, B. Am. Meteorol. Soc., 90, 1467–1485,https://doi.org/10.1175/2009BAMS2778.1, 2009.

Meehl, G. A., Goddard, L., Boer, G., Burgman, R., Branstator,G., Cassou, C., Corti, S., Danabasoglu, G., Doblas-Reyes, F.,Hawkins, E., Karspeck, A., Kimoto, M., Kumar, A., Matei, D.,Mignot, J., Msadek, R., Pohlmann, H., Rienecker, M., Rosati, T.,Schneider, E., Smith, D., Sutton, R., Teng, H., van Oldenborgh,G. J., Vecchi, G., and Yeager, S.: Decadal Climate Prediction: An


https://doi.org/10.1029/2012GL051644

https://doi.org/10.1002/grl.50256

https://doi.org/10.5194/bg-12-5211-2015

https://doi.org/10.5194/bg-12-5211-2015

https://doi.org/10.5194/gmd-8-2929-2015

https://doi.org/10.5194/bg-12-6955-2015

https://doi.org/10.1175/JCLI3898.1

https://doi.org/10.5194/essd-7-47-2015

https://doi.org/10.1029/2011GB004145


http://www.nemo-ocean.eu/About-NEMO/Reference-manuals

http://www.nemo-ocean.eu/About-NEMO/Reference-manuals

https://doi.org/10.1029/93GB01355

https://doi.org/10.1007/BF01054491

https://doi.org/10.1029/2007GB002958

https://doi.org/10.5194/tc-6-1383-2012


https://doi.org/10.1016/j.marchem.2015.04.005

https://doi.org/10.1175/2009BAMS2778.1


Update from the Trenches, B. Am. Meteorol. Soc., 95, 243–267,https://doi.org/10.1175/BAMS-D-12-00241.1, 2013.

Meehl, G. A., Moss, R., Taylor, K. E., Eyring, V., Stouffer, R.J., Bony, S., and Stevens, B.: Climate Model Intercomparisons:Preparing for the Next Phase, Eos Trans. AGU, 95, 77–78,https://doi.org/10.1002/2014EO090001, 2014.

Mignot, J., Swingedouw, D., Deshayes, J., Marti, O., Talandier, C.,Séférian, R., Lengaigne, M., and Madec, G.: On the evolution ofthe oceanic component of the IPSL climate models from CMIP3to CMIP5: A mean state comparison, Ocean Model., 72, 167–184, 2013.

Mikaloff Fletcher, S. E., Gruber, N., Jacobson, A. R., Gloor,M., Doney, S. C., Dutkiewicz, S., Gerber, M., Follows,M., Joos, F., Lindsay, K., Menemenlis, D., Mouchet, A.,Müller, S. A., and Sarmiento, J. L.: Inverse estimates of theoceanic sources and sinks of natural CO2 and the impliedoceanic carbon transport, Global Biogeochem. Cy., 21, GB1010,https://doi.org/10.1029/2006GB002751, 2007.

Moore, J., Doney, S., Kleypas, J., Glover, D., and Fung, I.: An in-termediate complexity marine ecosystem model for the globaldomain, Deep-Sea Res. Pt. II, 49, 403–462, 2002.

Moore, J., Doney, S., and Lindsay, K.: Upper oceanecosystem dynamics and iron cycling in a global three-dimensional model, Global Biogeochem. Cy., 18, GB4028,https://doi.org/10.1029/2004GB002220, 2004.

Orr, J. C.: Global Ocean Storage of Anthropogenic Carbon(GOSAC), final report, EC Environ. and Clim. Programme, Inst.Pierre Simon Laplace, Paris, France, 117 pp., 2002.

Phillips, T. J., Potter, G. L., Williamson, D. L., Cederwall, R. T.,Boyle, J. S., Fiorino, M., Hnilo, J. J., Olson, J. G., Xie, S., andYio, J. J.: Evaluating Parameterizations in General CirculationModels: Climate Simulation Meets Weather Prediction, B. Am.Meteorol. Soc., 85, 1903–1915, https://doi.org/10.1175/BAMS-85-12-1903, 2004.

Resplandy, L., Bopp, L., Orr, J. C., and Dunne, J. P.: Role of modeand intermediate waters in future ocean acidification: Analysis ofCMIP5 models, Geophys. Res. Lett., 40, 3091–3095, 2013.

Resplandy, L., Séférian, R., and Bopp, L.: Natural variability ofCO2 and O2 fluxes: What can we learn from centuries-long cli-mate models simulations? J. Geophys. Res.-Oceans, 120, 384–404, https://doi.org/10.1002/2014JC010463, 2015.

Rodgers, K. B., Lin, J., and Frölicher, T. L.: Emergence ofmultiple ocean ecosystem drivers in a large ensemble suitewith an Earth system model, Biogeosciences, 12, 3301–3320,https://doi.org/10.5194/bg-12-3301-2015, 2015.

Romanou, A., Gregg, W. W., Romanski, J., Kelley, M., Bleck,R., Healy, R., Nazarenko, L., Russell, G., Schmidt, G. A.,Sun, S., and Tausnev, N.: Natural air-sea flux of CO2 in sim-ulations of the NASA-GISS climate model: Sensitivity to thephysical ocean model formulation. Ocean Model., 66, 26–44,https://doi.org/10.1016/j.ocemod.2013.01.008, 2013.

Romanou, A., Romanski, J., and Gregg, W. W.: Natural oceancarbon cycle sensitivity to parameterizations of the recy-cling in a climate model, Biogeosciences, 11, 1137–1154,https://doi.org/10.5194/bg-11-1137-2014, 2014.

Rose, K. A., Roth, B. M., and Smith, E. P.: Skill assessment of spa-tial maps for oceanographic modeling, J. Marine Syst., 76, 34–48, https://doi.org/10.1016/j.jmarsys.2008.05.013, 2009.

Roy, T., Bopp, L., Gehlen, M., Schneider, B., Cadule, P.,Frölicher, T. L., Segschneider, J., Tjiputra, J., Heinze, C.,and Joos, F.: Regional Impacts of Climate Change and At-mospheric CO2 on Future Ocean Carbon Uptake: A Multi-model Linear Feedback Analysis, J. Climate, 24, 2300–2318,https://doi.org/10.1175/2010JCLI3787.1, 2011.

Sarmiento, J. L. and Gruber, N.: Ocean Biogeochemical Dynamics,Princeton University Press, Princeton, New Jersey, USA, 526 pp.,2006.

Schwinger, J., Tjiputra, J. F., Heinze, C., Bopp, L., Christian,J. R., Gehlen, M., Ilyina, T., Jones, C. D., Salas-Mélia, D.,Segschneider, J., Séférian, R., and Totterdell, I.: Nonlinearity ofOcean Carbon Cycle Feedbacks in CMIP5 Earth System Mod-els, J. Climate, 27, 3869–3888, https://doi.org/10.1175/JCLI-D-13-00452.1, 2014.

Séférian, R., Iudicone, D., Bopp, L., Roy, T., and Madec, G.: Wa-ter Mass Analysis of Effect of Climate Change on Air–SeaCO2 Fluxes: The Southern Ocean, J. Climate, 25, 3894–3908,https://doi.org/10.1175/JCLI-D-11-00291.1, 2012.

Séférian, R., Bopp, L., Gehlen, M., Orr, J., Ethé, C., Cadule,P., Aumont, O., Salas y Mélia, D., Voldoire, A., and Madec,G.: Skill assessment of three earth system models with com-mon marine biogeochemistry, Clim. Dynam., 40, 2549–2573,https://doi.org/10.1007/s00382-012-1362-8, 2013.

Séférian, R., Ribes, A., and Bopp, L.: Detecting the an-thropogenic influences on recent changes in oceancarbon uptake, Geophys. Res. Lett., 41, 5968–5977,https://doi.org/10.1002/2014GL061223, 2014.

Séférian, R., Delire, C., Decharme, B., Voldoire, A., Salas y Melia,D., Chevallier, M., Saint-Martin, D., Aumont, O., Calvet, J.-C., Carrer, D., Douville, H., Franchistéguy, L., Joetzjer, E., andSénési, S.: Development and evaluation of CNRM Earth sys-tem model – CNRM-ESM1, Geosci. Model Dev., 9, 1423–1453,https://doi.org/10.5194/gmd-9-1423-2016, 2016.

Sen Gupta, A. S., Muir, L. C., Brown, J. N., Phipps, S. J.,Durack, P. J., Monselesan, D., and Wijffels, S. E.: ClimateDrift in the CMIP3 Models, J. Climate, 25, 4621–4640,https://doi.org/10.1175/JCLI-D-11-00312.1, 2012.

Sen Gupta, A. S., Jourdain, N. C., Brown, J. N., and Monselesan,D.: Climate Drift in the CMIP5 models, J. Climate, 26, 8597–8615, https://doi.org/10.1175/JCLI-D-12-00521.1, 2013.

Servonnat, J., Mignot, J., Guilyardi, E., Swingedouw, D.,Séférian, R., and Labetoulle, S.: Reconstructing the sub-surface ocean decadal variability using surface nudgingin a perfect model framework, Clim. Dynam., 44, 1–24,https://doi.org/10.1007/s00382-014-2184-7, 2014.

Smith, D. M., Cusack, S., Colman, A. W., Folland, C. K., Harris, G.R., and Murphy, J. M.: Improved Surface Temperature Predictionfor the Coming Decade from a Global Climate Model, Science,317, 796–799, https://doi.org/10.1126/science.1139540, 2007.

Smith, M. J., Palmer, P. I., Purves, D. W., Vanderwel, M. C.,Lyutsarev, V., Calderhead, B., Joppa, L. N., Bishop, C. M.,and Emmott, S.: Changing how Earth System Modelling isdone to provide more useful information for decision making,science and society, B. Am. Meteorol. Soc., 95, 1453–1464,https://doi.org/10.1175/BAMS-D-13-00080.1, 2014.

Steinacher, M., Joos, F., Frölicher, T. L., Bopp, L., Cadule, P.,Cocco, V., Doney, S. C., Gehlen, M., Lindsay, K., Moore, J. K.,Schneider, B., and Segschneider, J.: Projected 21st century de-


https://doi.org/10.1175/BAMS-D-12-00241.1

https://doi.org/10.1002/2014EO090001

https://doi.org/10.1029/2006GB002751

https://doi.org/10.1029/2004GB002220

https://doi.org/10.1175/BAMS-85-12-1903

https://doi.org/10.1175/BAMS-85-12-1903

https://doi.org/10.1002/2014JC010463

https://doi.org/10.5194/bg-12-3301-2015

https://doi.org/10.1016/j.ocemod.2013.01.008

https://doi.org/10.5194/bg-11-1137-2014


https://doi.org/10.1175/2010JCLI3787.1




https://doi.org/10.1007/s00382-012-1362-8

https://doi.org/10.1002/2014GL061223

https://doi.org/10.5194/gmd-9-1423-2016



https://doi.org/10.1007/s00382-014-2184-7


https://doi.org/10.1175/BAMS-D-13-00080.1


crease in marine productivity: a multi-model analysis, Biogeo-sciences, 7, 979–1005, https://doi.org/10.5194/bg-7-979-2010,2010.

Stouffer, R. J., Weaver, A. J., and Eby, M.: A method forobtaining pre-twentieth century initial conditions for usein climate change studies, Clim. Dynam., 23, 327–339,https://doi.org/10.1007/s00382-004-0446-5, 2004.

Stow, C. A., Jolliff, J., McGillicuddy, D. J. J., Doney, S.C., Allen, J. I., Friedrichs, M. A. M., Rose, K. A., andWallhead, P.: Skill assessment for coupled biological/physicalmodels of marine systems, J. Marine Syst., 76, 4–15,https://doi.org/10.1016/j.jmarsys.2008.03.011, 2009.

Swingedouw, D., Mignot, J., Labetoulle, S., Guilyardi, E., andMadec, G.: Initialisation and predictability of the AMOC overthe last 50 years in a climate model, Clim. Dynam., 40, 2381–2399, https://doi.org/10.1007/s00382-012-1516-8, 2013.

Tagliabue, A. and Völker, C.: Towards accounting for dissolvediron speciation in global ocean models, Biogeosciences, 8, 3025–3039, https://doi.org/10.5194/bg-8-3025-2011, 2011.

Tagliabue, A., Bopp, L., and Gehlen, M.: The responseof marine carbon and nutrient cycles to ocean acidifica-tion: Large uncertainties related to phytoplankton physio-logical assumptions, Global Biogeochem. Cy., 25, GB3017,https://doi.org/10.1029/2010GB003929, 2011.

Takahashi, T., Broecker, W., and Langer, S.: Redfield Ratio Basedon Chemical-Data From Isopycnal Surfaces, J. Geophys. Res.-Oceans, 90, 6907–6924, 1985.

Tanhua, T., Koertzinger, A., Friis, K., Waugh, D. W.,and Wallace, D. W. R.: An estimate of anthropogenicCO2 inventory from decadal changes in oceanic car-bon content, P. Natl. Acad. Sci. USA, 104, 3037–3042,https://doi.org/10.1073/pnas.0606574104, 2007.

Tegen, I. and Fung, I.: Contribution to the Atmospheric MineralAerosol Load From Land-Surface Modification, J. Geophys.Res.-Atmos., 100, 18707–18726, 1995.

Tjiputra, J. F., Roelandt, C., Bentsen, M., Lawrence, D. M.,Lorentzen, T., Schwinger, J., Seland, Ø., and Heinze, C.: Eval-uation of the carbon cycle components in the Norwegian EarthSystem Model (NorESM), Geosci. Model Dev., 6, 301–325,https://doi.org/10.5194/gmd-6-301-2013, 2013.

Tjiputra, J. F., Olsen, A., Bopp, L., Lenton, A., Pfeil, B., Roy, T.,Segschneider, J., Totterdell, I., and Heinze, C.: Long-term surfacepCO2 trends from observations and models, Tellus B, 66, 151–168, https://doi.org/10.1007/s00382-007-0342-x, 2014.

Vancoppenolle, M., Bopp, L., Madec, G., Dunne, J. P., Ilyina, T.,Halloran, P. R., and Steiner, N.: Future Arctic Ocean primaryproductivity from CMIP5 simulations: Uncertain outcome, butconsistent mechanisms, Global Biogeochem. Cy., 27, 605–619,2013.

Vichi, M., Manzini, E., Fogli, P. G., Alessandri, A., Patara, L.,Scoccimarro, E., Masina, S., and Navarra, A.: Global and re-gional ocean carbon uptake and climate change: sensitivity toa substantial mitigation scenario, Clim. Dynam., 37, 1929–1947,https://doi.org/10.1007/s00382-011-1079-0, 2011.

Volodin, E. M., Dianskii, N. A., and Gusev, A. V.: Sim-ulating present-day climate with the INMCM4.0 cou-pled model of the atmospheric and oceanic general cir-culations, Izv. Atmos. Ocean. Phy. +, 46, 414–431,https://doi.org/10.1134/S000143381004002X, 2010.

Walin, G., Hieronymus, J., and Nycander, J.: Source-relatedvariables for the description of the oceanic carbon sys-tem, Geochem. Geophys. Geosyst., 15, 3675–3687,https://doi.org/10.1002/2014GC005383, 2014.

Wanninkhof, R.: A relationship between wind speed and gas ex-change over the ocean, J. Geophys. Res., 97, 7373–7382, 1992.

Wassmann, P., Duarte, C. M., Agustí, S., and Sejr, M. K.: Foot-prints of climate change in the Arctic marine ecosystem, Glob.Change Biol., 17, 1235–1249, https://doi.org/10.1111/j.1365-2486.2010.02311.x, 2010.

Watanabe, S., Hajima, T., Sudo, K., Nagashima, T., Takemura, T.,Okajima, H., Nozawa, T., Kawase, H., Abe, M., Yokohata, T.,Ise, T., Sato, H., Kato, E., Takata, K., Emori, S., and Kawamiya,M.: MIROC-ESM 2010: model description and basic results ofCMIP5-20c3m experiments, Geosci. Model Dev., 4, 845–872,https://doi.org/10.5194/gmd-4-845-2011, 2011.

Wenzel, S., Cox, P. M., Eyring, V., and Friedlingstein, P.: Emer-gent constraints on climate-carbon cycle feedbacks in the CMIP5Earth system models, J. Geophys. Res.-Biogeosci., 119, 794–807, https://doi.org/10.1002/2013JG002591, 2014.

Wu, T., Li, W., Ji, J., Xin, X., Li, L., Wang, Z., Zhang, Y.,Li, J., Zhang, F., Wei, M., Shi, X., Wu, F., Zhang, L., Chu,M., Jie, W., Liu, Y., Wang, F., Liu, X., Li, Q., Dong, M.,Liang, X., Gao, Y., and Zhang, J.: Global carbon budgets sim-ulated by the Beijing Climate Center Climate System Modelfor the last century, J. Geophys. Res.-Atmos., 118, 4326–4347,https://doi.org/10.1002/jgrd.50320, 2013.

Wunsch, C. and Heimbach, P.: Practical globaloceanic state estimation, Physica D, 230, 197–208,https://doi.org/10.1016/j.physd.2006.09.040, 2007.

Wunsch, C. and Heimbach, P.: How long to oceanic tracerand proxy equilibrium?, Quaternary Sci. Rev., 27, 637–651,https://doi.org/10.1016/j.quascirev.2008.01.006, 2008.

Yool, A., Oschlies, A., Nurser, A. J. G., and Gruber, N.: A model-based assessment of the TrOCA approach for estimating an-thropogenic carbon in the ocean, Biogeosciences, 7, 723–751,https://doi.org/10.5194/bg-7-723-2010, 2010.

Yool, A., Popova, E. E., and Anderson, T. R.: MEDUSA-2.0: anintermediate complexity biogeochemical model of the marinecarbon cycle for climate change and ocean acidification studies,Geosci. Model Dev., 6, 1767–1811, https://doi.org/10.5194/gmd-6-1767-2013, 2013.

Zeebe, R. E. and Wolf-Gladrow, D. A.: CO2 in seawater: equilib-rium, kinetics, isotopes, Oceanography Book Series, vol. 65, El-sevier, Amsterdam, the Netherlands, 346 pp., 2001.


https://doi.org/10.5194/bg-7-979-2010

https://doi.org/10.1007/s00382-004-0446-5


https://doi.org/10.1007/s00382-012-1516-8

https://doi.org/10.5194/bg-8-3025-2011

https://doi.org/10.1029/2010GB003929

https://doi.org/10.1073/pnas.0606574104

https://doi.org/10.5194/gmd-6-301-2013

https://doi.org/10.1007/s00382-007-0342-x

https://doi.org/10.1007/s00382-011-1079-0

https://doi.org/10.1134/S000143381004002X

https://doi.org/10.1002/2014GC005383

https://doi.org/10.1111/j.1365-2486.2010.02311.x

https://doi.org/10.1111/j.1365-2486.2010.02311.x

https://doi.org/10.5194/gmd-4-845-2011

https://doi.org/10.1002/2013JG002591

https://doi.org/10.1002/jgrd.50320

https://doi.org/10.1016/j.physd.2006.09.040

https://doi.org/10.1016/j.quascirev.2008.01.006

https://doi.org/10.5194/bg-7-723-2010

https://doi.org/10.5194/gmd-6-1767-2013

https://doi.org/10.5194/gmd-6-1767-2013

inconsistent strategies to spin up models in cmip5 ... · rent generation biogeochemical models...

Documents